{ "name": "LLM World — tracked large language model releases", "homepage": "https://mayoorrnikam.github.io/llm-world/", "schema_version": "1.6", "updated": "2026-10-15", "count": 280, "license": "CC-BY-4.0", "license_url": "https://creativecommons.org/licenses/by/4.0/", "attribution": "Release dates and metadata from LLM World — https://mayoorrnikam.github.io/llm-world/ — CC BY 4.0", "notice": "Undisclosed figures are null, never estimated. Check each record's sources[] before relying on a figure; provenance.status records how well it was verified.", "releases": [ { "id": "palm", "model": "PaLM", "company": "Google DeepMind", "family": "PaLM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding", "multilingual" ], "tags": [ "flagship" ], "note": "Google's 540-billion-parameter model that demonstrated breakthrough few-shot scaling.", "events": [ { "type": "announcement", "date": "2022-04-04", "sources": [ "palm-s1", "palm-s2" ] } ], "specifications": { "language": { "context_window": 2048, "parameter_count": 540000000000 } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "palm-s1", "url": "https://research.google/blog/pathways-language-model-palm-scaling-to-540-billion-parameters-for-breakthrough-performance/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240412114709/https://research.google/blog/pathways-language-model-palm-scaling-to-540-billion-parameters-for-breakthrough-performance/", "retrieved": "2024-04-12" }, { "id": "palm-s2", "url": "https://arxiv.org/abs/2204.02311", "type": "technical_paper", "authority": "primary", "archived_url": "https://web.archive.org/web/20220406011033/https://arxiv.org/abs/2204.02311", "retrieved": "2022-04-06" }, { "id": "palm-paper-pdf", "url": "https://arxiv.org/pdf/2204.02311", "type": "technical_paper", "authority": "primary", "archived_url": "https://web.archive.org/web/20240522140427/https://arxiv.org/pdf/2204.02311", "retrieved": "2024-05-22" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement and the technical paper. Not verified because the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Capabilities evidenced from the primary sources: multilingual: “PaLM was trained using a combination of English and multilingual datasets that include high-quality web documents, books, Wikipedia, conversations, and GitHub code.” | coding: “PaLM 540B shows strong performance across coding tasks and natural language tasks in a single model, even though it has only 5% code in the pre-training dataset.”." }, "evidence": { "release_date": [ { "value": "2022-04-04", "sources": [ "palm-s1" ] } ], "context_window": [ { "value": 2048, "sources": [ "palm-paper-pdf" ] } ], "parameter_count": [ { "value": 540000000000, "sources": [ "palm-s1", "palm-s2", "palm-paper-pdf" ] } ] } }, { "id": "opt-175b", "model": "OPT-175B", "company": "Meta AI", "family": "OPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "Meta's open 175-billion-parameter model released to democratize access to large language models.", "events": [ { "type": "announcement", "date": "2022-05-05", "sources": [ "opt-175b-s1", "opt-175b-s2" ] } ], "specifications": { "language": { "context_window": 2048, "parameter_count": 175000000000 } }, "access": { "open_weights": true, "license": "OPT-175B License Agreement" }, "sources": [ { "id": "opt-175b-s1", "url": "https://arxiv.org/abs/2205.01068", "type": "technical_paper", "authority": "primary", "archived_url": "https://web.archive.org/web/20220504144732/https://arxiv.org/abs/2205.01068", "retrieved": "2022-05-04" }, { "id": "opt-175b-s2", "url": "https://github.com/facebookresearch/metaseq", "type": "official_repository", "authority": "primary", "archived_url": "https://web.archive.org/web/20220504144833/https://github.com/facebookresearch/metaseq", "retrieved": "2022-05-04" } ], "provenance": { "status": "estimated", "confidence": 50, "reason": "Cited to the technical paper and the official repository. Not verified because the release date and the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present." }, "evidence": { "parameter_count": [ { "value": 175000000000, "sources": [ "opt-175b-s1", "opt-175b-s2" ] } ] } }, { "id": "bloom", "model": "BLOOM", "company": "BigScience", "family": "BLOOM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "multilingual" ], "tags": [], "note": "The 176-billion-parameter multilingual open model built by over a thousand volunteer researchers.", "events": [ { "type": "announcement", "date": "2022-07-12", "sources": [ "bloom-s1" ] } ], "specifications": { "language": { "context_window": 2048, "parameter_count": 176247271424 } }, "access": { "open_weights": true, "license": "BigScience BLOOM RAIL 1.0" }, "sources": [ { "id": "bloom-s1", "url": "https://huggingface.co/bigscience/bloom", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20220711092607/https://huggingface.co/bigscience/bloom", "retrieved": "2022-07-11" } ], "provenance": { "status": "partially_verified", "confidence": 75, "reason": "Cited to the official model card. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Capabilities evidenced from the primary sources: multilingual: “0 bloom feature-extraction Model card Files Files and versions Community 23 Train Deploy Use in Transformers Edit model card BLOOM LM BigScience Large Open-science Open-a”." }, "evidence": { "context_window": [ { "value": 2048, "sources": [ "bloom-s1" ] } ], "parameter_count": [ { "value": 176247271424, "sources": [ "bloom-s1" ] } ] } }, { "id": "glm-130b", "model": "GLM-130B", "company": "Zhipu AI", "family": "GLM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "multilingual" ], "tags": [], "note": "China's landmark open bilingual 130-billion-parameter model released in August 2022.", "events": [ { "type": "announcement", "date": "2022-08-01", "sources": [ "glm-130b-s1" ] } ], "specifications": { "language": { "context_window": 2048, "parameter_count": 130000000000 } }, "access": { "open_weights": true, "license": "The GLM-130B License" }, "sources": [ { "id": "glm-130b-s1", "url": "https://github.com/zai-org/GLM-130B", "type": "official_repository", "authority": "primary", "archived_url": "https://web.archive.org/web/20250729052130/https://github.com/zai-org/GLM-130B", "retrieved": "2025-07-29" } ], "provenance": { "status": "estimated", "confidence": 50, "reason": "Cited to the official repository. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Capabilities evidenced from the primary sources: multilingual: “For smaller models, please find monolingual GLMs (English: 10B/2B/515M/410M/335M/110M, Chinese: 10B/335M) and an 1B multilingual GLM (104 languages).”." }, "evidence": { "context_window": [ { "value": 2048, "sources": [ "glm-130b-s1" ] } ], "parameter_count": [ { "value": 130000000000, "sources": [ "glm-130b-s1" ] } ] } }, { "id": "whisper", "model": "Whisper", "company": "OpenAI", "family": "Whisper", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [ "audio", "multilingual" ], "tags": [], "note": "OpenAI's speech recognition model, released with open weights and inference code.", "events": [ { "type": "announcement", "date": "2022-09-21", "sources": [ "whisper-s1" ] } ], "specifications": {}, "access": { "open_weights": true, "license": null }, "sources": [ { "id": "whisper-s1", "url": "https://openai.com/index/whisper/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240502131713/https://openai.com/index/whisper", "retrieved": "2024-05-02" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Capabilities evidenced from the primary sources: audio: “static [&>div]:h-full\"> Research Products Safety Company September 21, 2022 Introducing Whisper We’ve trained and are open-sourcing a neural net called Whisper that approaches huma” | multilingual: “Whisper is an automatic speech recognition (ASR) system trained on 680,000 hours of multilingual and multitask supervised data collected from the web.”." }, "evidence": { "release_date": [ { "value": "2022-09-21", "sources": [ "whisper-s1" ] } ] }, "undisclosed": [ "license" ] }, { "id": "llama-1", "model": "Llama 1", "company": "Meta AI", "family": "Llama", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "Open foundation models from 7B to 65B parameters that ignited the open-source LLM ecosystem.", "events": [ { "type": "announcement", "date": "2023-02-24", "sources": [ "llama-1-s1" ] } ], "specifications": { "language": { "context_window": 2048, "parameter_count": 65000000000 } }, "access": { "open_weights": true, "license": "LLaMA Non-Commercial Research License" }, "sources": [ { "id": "llama-1-s1", "url": "https://ai.meta.com/blog/large-language-model-llama-meta-ai/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20230714040519/https://ai.meta.com/blog/large-language-model-llama-meta-ai/", "retrieved": "2023-07-14" } ], "provenance": { "status": "partially_verified", "confidence": 75, "reason": "Cited to the lab's own announcement. Not verified because the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": { "release_date": [ { "value": "2023-02-24", "sources": [ "llama-1-s1" ] } ], "parameter_count": [ { "value": 65000000000, "sources": [ "llama-1-s1" ] } ] } }, { "id": "claude-1", "model": "Claude 1", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding" ], "tags": [ "flagship" ], "note": "Anthropic's debut Claude model focused on helpfulness, harmlessness, and long-context dialogue.", "events": [ { "type": "announcement", "date": "2023-03-14", "sources": [ "claude-1-s1", "claude-1-s2" ] } ], "specifications": { "language": { "context_window": 9000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-1-s1", "url": "https://www.anthropic.com/news/introducing-claude", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240117131957/https://www.anthropic.com/news/introducing-claude", "retrieved": "2024-01-17" }, { "id": "claude-1-s2", "url": "https://www.anthropic.com/news/100k-context-windows", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240118225720/https://www.anthropic.com/news/100k-context-windows", "retrieved": "2024-01-18" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (Mar 14, 2023) and context window (9K) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Capabilities evidenced from the primary sources: coding: “Claude can help with use cases including summarization, search, creative and collaborative writing, Q&A, coding, and more.”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2023-03-14", "sources": [ "claude-1-s1" ] } ], "context_window": [ { "value": 9000, "sources": [ "claude-1-s2" ] } ] } }, { "id": "gpt-4", "model": "GPT-4", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision" ], "tags": [ "flagship" ], "note": "OpenAI's first multimodal flagship that set new standards across professional and academic benchmarks.", "events": [ { "type": "announcement", "date": "2023-03-14", "sources": [ "gpt-4-s1", "gpt-4-s2" ] } ], "specifications": { "language": { "context_window": 8192, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-4-s1", "url": "https://openai.com/index/gpt-4-research/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240502194322/https://openai.com/index/gpt-4-research", "retrieved": "2024-05-02" }, { "id": "gpt-4-s2", "url": "https://developers.openai.com/api/docs/models/gpt-4", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260222002942/https://developers.openai.com/api/docs/models/gpt-4", "retrieved": "2026-02-22" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (March 14, 2023) and context window (8,192) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Capabilities evidenced from the primary sources: vision: “GPT-4 is a large multimodal model (accepting image and text inputs, emitting text outputs) that, while less capable than humans in many real-world scenarios, exhibits hum”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2023-03-14", "sources": [ "gpt-4-s1" ] } ], "context_window": [ { "value": 8192, "sources": [ "gpt-4-s1", "gpt-4-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 30, "output": 60 }, "currency": "USD", "sources": [ "gpt-4-s2" ], "observed_on": "2026-02-22" } ] }, { "id": "palm-2", "model": "PaLM 2", "company": "Google DeepMind", "family": "PaLM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding", "multilingual" ], "tags": [ "flagship" ], "note": "Google's next-generation model family that powered Bard and more than 25 other products.", "events": [ { "type": "announcement", "date": "2023-05-10", "sources": [ "palm-2-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "palm-2-s1", "url": "https://blog.google/technology/ai/google-palm-2-ai-large-language-model/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20230510180653/https://blog.google/technology/ai/google-palm-2-ai-large-language-model/", "retrieved": "2023-05-10" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (May 10, 2023) — was found in an archived primary source, checked 2026-08-10. Context window and parameter count are not recorded, so nothing is claimed about them. Capabilities evidenced from the primary sources: coding: “PaLM 2 is a state-of-the-art language model with improved multilingual, reasoning and coding capabilities.” | multilingual: “PaLM 2 is a state-of-the-art language model with improved multilingual, reasoning and coding capabilities.”." }, "undisclosed": [ "context_window", "parameter_count" ], "evidence": { "release_date": [ { "value": "2023-05-10", "sources": [ "palm-2-s1" ] } ] } }, { "id": "claude-2", "model": "Claude 2", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding" ], "tags": [ "flagship" ], "note": "Doubled Claude's context window to 100K tokens with stronger coding and writing abilities.", "events": [ { "type": "announcement", "date": "2023-07-11", "sources": [ "claude-2-s1" ] } ], "specifications": { "language": { "context_window": 100000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-2-s1", "url": "https://www.anthropic.com/news/claude-2", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240117223805/https://www.anthropic.com/news/claude-2", "retrieved": "2024-01-17" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (Jul 11, 2023) and context window (100K) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present. Capabilities evidenced from the primary sources: coding: “Their coding assistant Cody uses Claude 2’s improved reasoning ability to give even more accurate answers to user queries while also passing along more codebase context w”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2023-07-11", "sources": [ "claude-2-s1" ] } ], "context_window": [ { "value": 100000, "sources": [ "claude-2-s1" ] } ] } }, { "id": "llama-2", "model": "Llama 2", "company": "Meta AI", "family": "Llama", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "The first major open model with commercial licensing, defining open-source AI for a year.", "events": [ { "type": "announcement", "date": "2023-07-18", "sources": [ "llama-2-s1", "llama-2-s2" ] } ], "specifications": { "language": { "context_window": 4096, "parameter_count": 70000000000 } }, "access": { "open_weights": true, "license": "Llama 2 Community License" }, "sources": [ { "id": "llama-2-s1", "url": "https://ai.meta.com/blog/llama-2/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20230719035227/https://ai.meta.com/blog/llama-2/", "retrieved": "2023-07-19" }, { "id": "llama-2-s2", "url": "https://huggingface.co/meta-llama/Llama-2-70b-hf", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20230718195233/https://huggingface.co/meta-llama/Llama-2-70b-hf", "retrieved": "2023-07-18" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement and the official model card. Not verified because the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. The primary source states the model is text-only." }, "evidence": { "release_date": [ { "value": "2023-07-18", "sources": [ "llama-2-s1" ] } ], "context_window": [ { "value": 4096, "sources": [ "llama-2-s2" ] } ], "parameter_count": [ { "value": 70000000000, "sources": [ "llama-2-s2" ] } ] } }, { "id": "mistral-7b", "model": "Mistral 7B", "company": "Mistral AI", "family": "Mistral", "classification": { "primary_type": "language", "subtype": "slm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding" ], "tags": [ "small-efficient" ], "note": "A 7-billion-parameter model outperforming much larger models and putting European open AI on the map.", "events": [ { "type": "announcement", "date": "2023-09-27", "sources": [ "mistral-7b-s1", "mistral-7b-s2", "mistral-7b-s3" ] } ], "specifications": { "language": { "context_window": 8192, "parameter_count": 7300000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "mistral-7b-s1", "url": "https://mistral.ai/news/announcing-mistral-7b/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20230927145256/https://mistral.ai/news/announcing-mistral-7b/", "retrieved": "2023-09-27" }, { "id": "mistral-7b-s2", "url": "https://mistral.ai/news/announcing-mistral-7b", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20230927145256/https://mistral.ai/news/announcing-mistral-7b/", "retrieved": "2023-09-27" }, { "id": "mistral-7b-s3", "url": "https://huggingface.co/mistralai/Mistral-7B-v0.1", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20230927171925/https://huggingface.co/mistralai/Mistral-7B-v0.1", "retrieved": "2023-09-27" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (September 27, 2023), context window (8192) and parameter count (7.3B) — was found in an archived primary source, checked 2026-08-10. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present." }, "evidence": { "release_date": [ { "value": "2023-09-27", "sources": [ "mistral-7b-s1", "mistral-7b-s2" ] } ], "context_window": [ { "value": 8192, "sources": [ "mistral-7b-s1", "mistral-7b-s2" ] } ], "parameter_count": [ { "value": 7300000000, "sources": [ "mistral-7b-s1", "mistral-7b-s2" ] } ] } }, { "id": "grok-1", "model": "Grok-1", "company": "xAI", "family": "Grok", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding" ], "tags": [ "flagship" ], "note": "xAI's debut model with real-time knowledge of the X platform, released to Premium+ subscribers.", "events": [ { "type": "announcement", "date": "2023-11-03", "sources": [ "grok-1-s1", "grok-1-s2" ] }, { "type": "weights_availability", "date": "2024-03-17", "sources": [ "grok-1-open-s1", "grok-1-s2" ] } ], "specifications": { "language": { "context_window": 8192, "parameter_count": 314000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "grok-1-s1", "url": "https://x.ai/news/grok", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250314194650/https://x.ai/news/grok", "retrieved": "2025-03-14" }, { "id": "grok-1-s2", "url": "https://github.com/xai-org/grok-1", "type": "official_repository", "authority": "primary", "archived_url": "https://web.archive.org/web/20240317205046/https://github.com/xai-org/grok-1", "retrieved": "2024-03-17" }, { "id": "grok-1-open-s1", "url": "https://x.ai/news/grok-os", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250316195647/https://x.ai/news/grok-os", "retrieved": "2025-03-16" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement and the official repository. Not verified because the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "previous_ids": [ "grok-1-open" ], "evidence": { "release_date": [ { "value": "2023-11-03", "sources": [ "grok-1-s1" ] } ], "context_window": [ { "value": 8192, "sources": [ "grok-1-s1" ] } ], "parameter_count": [ { "value": 314000000000, "sources": [ "grok-1-s2", "grok-1-open-s1" ] } ] }, "benchmarks": [ { "name": "HumanEval", "score": 63.2, "evaluation_type": "vendor_reported", "reported_on": "2023-11-04", "sources": [ "grok-1-s1" ] }, { "name": "MMLU", "score": 73, "evaluation_type": "vendor_reported", "reported_on": "2023-11-04", "sources": [ "grok-1-s1" ] } ] }, { "id": "gemini-1-ultra", "model": "Gemini 1.0 Ultra", "company": "Google DeepMind", "family": "Gemini", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "capabilities": [ "vision", "audio" ], "tags": [ "flagship" ], "note": "Google's largest first-generation Gemini model, for highly complex tasks.", "events": [ { "type": "announcement", "date": "2023-12-06", "sources": [ "gemini-1-s1", "gemini-1-s2" ] } ], "specifications": { "language": { "context_window": 32768, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-1-s1", "url": "https://blog.google/technology/ai/google-gemini-ai/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20231206150851/https://blog.google/technology/ai/google-gemini-ai/", "retrieved": "2023-12-06" }, { "id": "gemini-1-s2", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_1_report.pdf", "type": "technical_paper", "authority": "primary", "archived_url": "https://web.archive.org/web/20231206151650/https://storage.googleapis.com/deepmind-media/gemini/gemini_1_report.pdf", "retrieved": "2023-12-06" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Split from the former \"gemini-1\" record, which was named for the family while holding one variant's figures. Evidence: \"Gemini Ultra — our largest and most capable model for highly complex tasks.\" Capabilities evidenced from the primary sources: audio: “From natural image, audio and video understanding to mathematical reasoning, Gemini Ultra’s performance exceeds current state-of-the-art results on 30 of the 32 widely-us” | vision: “This highlights Gemini Ultra's n ati v e m ultim oda lit y and eludes t o its m ore co mpl ex rea so ning a biliti es acro ss interl ea v ed sequ en ces o f image and tex”." }, "evidence": { "release_date": [ { "value": "2023-12-06", "sources": [ "gemini-1-s1" ] } ], "context_window": [ { "value": 32768, "sources": [ "gemini-1-s2" ] } ] }, "undisclosed": [ "parameter_count" ] }, { "id": "gemini-1-pro", "model": "Gemini 1.0 Pro", "company": "Google DeepMind", "family": "Gemini", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "capabilities": [ "vision", "audio" ], "tags": [ "flagship" ], "note": "The first-generation Gemini model that shipped across Google products and the Gemini API.", "events": [ { "type": "announcement", "date": "2023-12-06", "sources": [ "gemini-1-s1", "gemini-1-s2" ] } ], "specifications": { "language": { "context_window": 32768, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-1-s1", "url": "https://blog.google/technology/ai/google-gemini-ai/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20231206150851/https://blog.google/technology/ai/google-gemini-ai/", "retrieved": "2023-12-06" }, { "id": "gemini-1-s2", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_1_report.pdf", "type": "technical_paper", "authority": "primary", "archived_url": "https://web.archive.org/web/20231206151650/https://storage.googleapis.com/deepmind-media/gemini/gemini_1_report.pdf", "retrieved": "2023-12-06" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Split from the former \"gemini-1\" record, which was named for the family while holding one variant's figures. Evidence: \"Gemini Pro — our best model for scaling across a wide range of tasks. The 32,768-token context is carried from the former family record; the announcement does not state per-model context windows, so it is recorded here only and will show as untraceable until a source states it.\" Capabilities evidenced from the primary sources: audio: “From natural image, audio and video understanding to mathematical reasoning, Gemini Ultra’s performance exceeds current state-of-the-art results on 30 of the 32 widely-us” | vision: “This highlights Gemini Ultra's n ati v e m ultim oda lit y and eludes t o its m ore co mpl ex rea so ning a biliti es acro ss interl ea v ed sequ en ces o f image and tex”." }, "evidence": { "release_date": [ { "value": "2023-12-06", "sources": [ "gemini-1-s1" ] } ], "context_window": [ { "value": 32768, "sources": [ "gemini-1-s2" ] } ] }, "undisclosed": [ "parameter_count" ] }, { "id": "gemini-1-nano", "model": "Gemini 1.0 Nano", "company": "Google DeepMind", "family": "Gemini", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "capabilities": [ "vision", "audio" ], "tags": [ "flagship" ], "note": "Google's on-device Gemini model, trained in 1.8B and 3.25B versions for low- and high-memory devices.", "events": [ { "type": "announcement", "date": "2023-12-06", "sources": [ "gemini-1-s1", "gemini-1-s2" ] } ], "specifications": { "language": { "context_window": 32768, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-1-s1", "url": "https://blog.google/technology/ai/google-gemini-ai/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20231206150851/https://blog.google/technology/ai/google-gemini-ai/", "retrieved": "2023-12-06" }, { "id": "gemini-1-s2", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_1_report.pdf", "type": "technical_paper", "authority": "primary", "archived_url": "https://web.archive.org/web/20231206151650/https://storage.googleapis.com/deepmind-media/gemini/gemini_1_report.pdf", "retrieved": "2023-12-06" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Split from the former \"gemini-1\" record, which was named for the family while holding one variant's figures. Evidence: \"Gemini Nano, our most efficient model for on-device tasks, via AICore, a new system capability available in Android 14, starting on Pixel 8 Pro devices.\" Capabilities evidenced from the primary sources: audio: “From natural image, audio and video understanding to mathematical reasoning, Gemini Ultra’s performance exceeds current state-of-the-art results on 30 of the 32 widely-us” | vision: “This highlights Gemini Ultra's n ati v e m ultim oda lit y and eludes t o its m ore co mpl ex rea so ning a biliti es acro ss interl ea v ed sequ en ces o f image and tex”." }, "evidence": { "release_date": [ { "value": "2023-12-06", "sources": [ "gemini-1-s1" ] } ], "context_window": [ { "value": 32768, "sources": [ "gemini-1-s2" ] } ] }, "undisclosed": [ "parameter_count" ] }, { "id": "mixtral", "model": "Mixtral 8x7B", "company": "Mistral AI", "family": "Mixtral", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding" ], "tags": [], "note": "The first open mixture-of-experts model to rival much larger dense models.", "events": [ { "type": "announcement", "date": "2023-12-11", "sources": [ "mixtral-s1", "mixtral-s2" ] } ], "specifications": { "language": { "context_window": 32768, "parameter_count": 46700000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "mixtral-s1", "url": "https://mistral.ai/news/mixtral-of-experts/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20231211075956/https://mistral.ai/news/mixtral-of-experts/", "retrieved": "2023-12-11" }, { "id": "mixtral-s2", "url": "https://mistral.ai/news/mixtral-of-experts", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20231211075956/https://mistral.ai/news/mixtral-of-experts/", "retrieved": "2023-12-11" } ], "provenance": { "status": "partially_verified", "confidence": 75, "reason": "Cited to the lab's own announcement. Not verified because the context window and the parameter count could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present." }, "evidence": { "release_date": [ { "value": "2023-12-11", "sources": [ "mixtral-s1", "mixtral-s2" ] } ], "context_window": [ { "value": 32768, "sources": [ "mixtral-s1", "mixtral-s2" ] } ] } }, { "id": "phi-2", "model": "Phi-2", "company": "Microsoft", "family": "Phi", "classification": { "primary_type": "language", "subtype": "slm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [ "small-efficient" ], "note": "A 2.7-billion-parameter model showing that small models can compete through high-quality curated data.", "events": [ { "type": "announcement", "date": "2023-12-12", "sources": [ "phi-2-s1", "phi-2-s2" ] } ], "specifications": { "language": { "context_window": 2048, "parameter_count": 2700000000 } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "phi-2-s1", "url": "https://www.microsoft.com/en-us/research/blog/phi-2-the-surprising-power-of-small-language-models/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20231212183225/https://www.microsoft.com/en-us/research/blog/phi-2-the-surprising-power-of-small-language-models/", "retrieved": "2023-12-12" }, { "id": "phi-2-s2", "url": "https://huggingface.co/microsoft/phi-2", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20231214040409/https://huggingface.co/microsoft/phi-2", "retrieved": "2023-12-14" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (December 12, 2023), context window (2048) and parameter count (2.7B) — was found in an archived primary source, checked 2026-08-10." }, "evidence": { "release_date": [ { "value": "2023-12-12", "sources": [ "phi-2-s1" ] } ], "context_window": [ { "value": 2048, "sources": [ "phi-2-s2" ] } ], "parameter_count": [ { "value": 2700000000, "sources": [ "phi-2-s1", "phi-2-s2" ] } ] } }, { "id": "glm-4", "model": "GLM-4", "company": "Zhipu AI", "family": "GLM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "multilingual" ], "tags": [ "flagship" ], "note": "Zhipu's flagship of the GLM-4 series, served through its API.", "events": [ { "type": "announcement", "date": "2024-01-16", "sources": [ "glm-4-s1", "glm-4-s2" ] } ], "specifications": { "language": { "context_window": 128000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "glm-4-s1", "url": "https://github.com/zai-org/GLM-4", "type": "official_repository", "authority": "primary", "archived_url": "https://web.archive.org/web/20250729052117/https://github.com/zai-org/GLM-4", "retrieved": "2025-07-29" }, { "id": "glm-4-s2", "url": "https://arxiv.org/abs/2406.12793", "type": "technical_paper", "authority": "primary", "archived_url": "https://web.archive.org/web/20240619020423/https://arxiv.org/abs/2406.12793", "retrieved": "2024-06-19" } ], "provenance": { "status": "partially_verified", "confidence": 75, "reason": "Split from the former \"glm-4\" record, which was named for the family while holding one variant's figures. Evidence: \"This report primarily focuses on the GLM-4 language series, which includes GLM-4, GLM-4-Air, and GLM-4-9B.\" Capabilities evidenced from the primary sources: multilingual: “7k GLM-4 series: Open Multilingual Multimodal Chat LMs | 开源多语言多模态对话模型 License Apache-2.”." }, "evidence": { "context_window": [ { "value": 128000, "sources": [ "glm-4-s1", "glm-4-s2" ] } ] } }, { "id": "glm-4-air", "model": "GLM-4-Air", "company": "Zhipu AI", "family": "GLM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "multilingual" ], "tags": [], "note": "The cost-efficient tier of the GLM-4 series, served through the API alongside the flagship.", "events": [ { "type": "announcement", "date": "2024-01-16", "sources": [ "glm-4-s1", "glm-4-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "glm-4-s1", "url": "https://github.com/zai-org/GLM-4", "type": "official_repository", "authority": "primary", "archived_url": "https://web.archive.org/web/20250729052117/https://github.com/zai-org/GLM-4", "retrieved": "2025-07-29" }, { "id": "glm-4-s2", "url": "https://arxiv.org/abs/2406.12793", "type": "technical_paper", "authority": "primary", "archived_url": "https://web.archive.org/web/20240619020423/https://arxiv.org/abs/2406.12793", "retrieved": "2024-06-19" } ], "provenance": { "status": "partially_verified", "confidence": 75, "reason": "Split from the former \"glm-4\" record, which was named for the family while holding one variant's figures. Evidence: \"This report primarily focuses on the GLM-4 language series, which includes GLM-4, GLM-4-Air, and GLM-4-9B.\" Capabilities evidenced from the primary sources: multilingual: “7k GLM-4 series: Open Multilingual Multimodal Chat LMs | 开源多语言多模态对话模型 License Apache-2.”." }, "evidence": {}, "undisclosed": [ "context_window" ] }, { "id": "glm-4-9b", "model": "GLM-4-9B", "company": "Zhipu AI", "family": "GLM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "multilingual" ], "tags": [], "note": "The open-weights member of the GLM-4 series, released with the ChatGLM technical report.", "events": [ { "type": "announcement", "date": "2024-01-16", "sources": [ "glm-4-s1", "glm-4-s2" ] } ], "specifications": { "language": { "context_window": 128000, "parameter_count": 9000000000 } }, "access": { "open_weights": true, "license": null }, "sources": [ { "id": "glm-4-s1", "url": "https://github.com/zai-org/GLM-4", "type": "official_repository", "authority": "primary", "archived_url": "https://web.archive.org/web/20250729052117/https://github.com/zai-org/GLM-4", "retrieved": "2025-07-29" }, { "id": "glm-4-s2", "url": "https://arxiv.org/abs/2406.12793", "type": "technical_paper", "authority": "primary", "archived_url": "https://web.archive.org/web/20240619020423/https://arxiv.org/abs/2406.12793", "retrieved": "2024-06-19" } ], "provenance": { "status": "partially_verified", "confidence": 75, "reason": "Split from the former \"glm-4\" record, which was named for the family while holding one variant's figures. Evidence: \"This report primarily focuses on the GLM-4 language series, which includes GLM-4, GLM-4-Air, and GLM-4-9B.\" Capabilities evidenced from the primary sources: multilingual: “7k GLM-4 series: Open Multilingual Multimodal Chat LMs | 开源多语言多模态对话模型 License Apache-2.”." }, "evidence": { "context_window": [ { "value": 128000, "sources": [ "glm-4-s1", "glm-4-s2" ] } ], "parameter_count": [ { "value": 9000000000, "sources": [ "glm-4-s1", "glm-4-s2" ] } ] } }, { "id": "qwen-vl-max", "model": "Qwen-VL-Max", "company": "Alibaba Qwen", "family": "Qwen-VL", "classification": { "primary_type": "vision", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "The API-only top of the Qwen-VL line, announced alongside Qwen-VL-Plus as an upgrade over the open Qwen-VL.", "events": [ { "type": "announcement", "date": "2024-01-25", "sources": [ "qwen-vl-max-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "qwen-vl-max-s1", "url": "https://qwenlm.github.io/blog/qwen-vl/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240130032159/https://qwenlm.github.io/blog/qwen-vl/", "retrieved": "2024-01-30" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2024-01-25", "sources": [ "qwen-vl-max-s1" ] } ] } }, { "id": "qwen1-5-72b", "model": "Qwen1.5-72B", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "The largest of the eight Qwen1.5 sizes, and the one the lab recommended; every size supports a 32,768-token context.", "events": [ { "type": "announcement", "date": "2024-02-04", "sources": [ "qwen1-5-72b-s1" ] } ], "specifications": { "language": { "context_window": 32768, "parameter_count": 72000000000 } }, "access": { "open_weights": true, "license": null }, "sources": [ { "id": "qwen1-5-72b-s1", "url": "https://qwenlm.github.io/blog/qwen1.5/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240205162344/https://qwenlm.github.io/blog/qwen1.5/", "retrieved": "2024-02-05" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present." }, "evidence": { "release_date": [ { "value": "2024-02-04", "sources": [ "qwen1-5-72b-s1" ] } ], "context_window": [ { "value": 32768, "sources": [ "qwen1-5-72b-s1" ] } ], "parameter_count": [ { "value": 72000000000, "sources": [ "qwen1-5-72b-s1" ] } ] }, "undisclosed": [ "license" ] }, { "id": "gemini-1-5-pro", "model": "Gemini 1.5 Pro", "company": "Google DeepMind", "family": "Gemini", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": [ "vision", "video" ], "tags": [ "flagship" ], "note": "Google's long-context MoE flagship that extended the context window to one million tokens.", "events": [ { "type": "announcement", "date": "2024-02-15", "sources": [ "gemini-1-5-pro-s1", "gemini-1-5-pro-s2" ] } ], "specifications": { "language": { "context_window": 2097152, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-1-5-pro-s1", "url": "https://blog.google/technology/ai/google-gemini-next-generation-model-february-2024/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240215151350/https://blog.google/technology/ai/google-gemini-next-generation-model-february-2024/", "retrieved": "2024-02-15" }, { "id": "gemini-1-5-pro-s2", "url": "https://developers.googleblog.com/en/new-features-for-the-gemini-api-and-google-ai-studio/", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20240627135644/https://developers.googleblog.com/en/new-features-for-the-gemini-api-and-google-ai-studio/", "retrieved": "2024-06-27" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement and official documentation. Not verified because the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2024-02-15", "sources": [ "gemini-1-5-pro-s1" ] } ], "context_window": [ { "value": 2097152, "sources": [ "gemini-1-5-pro-s2" ] } ] } }, { "id": "gemma", "model": "Gemma", "company": "Google DeepMind", "family": "Gemma", "classification": { "primary_type": "language", "subtype": "slm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [ "small-efficient" ], "note": "Google's first open-weight model family built from Gemini research for developers.", "events": [ { "type": "announcement", "date": "2024-02-21", "sources": [ "gemma-s1", "gemma-s2", "gemma-s3" ] } ], "specifications": { "language": { "context_window": 8192, "parameter_count": 7000000000 } }, "access": { "open_weights": true, "license": "Gemma Terms of Use" }, "sources": [ { "id": "gemma-s1", "url": "https://blog.google/technology/developers/gemma-open-models/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240221130902/https://blog.google/technology/developers/gemma-open-models/", "retrieved": "2024-02-21" }, { "id": "gemma-s2", "url": "https://huggingface.co/google/gemma-7b", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20240221145741/https://huggingface.co/google/gemma-7b", "retrieved": "2024-02-21" }, { "id": "gemma-s3", "url": "https://ai.google.dev/gemma/terms", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240222072002/https://ai.google.dev/gemma/terms", "retrieved": "2024-02-22" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement and the official model card. Not verified because the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": { "release_date": [ { "value": "2024-02-21", "sources": [ "gemma-s1", "gemma-s2", "gemma-s3" ] } ], "context_window": [ { "value": 8192, "sources": [ "gemma-s2" ] } ], "parameter_count": [ { "value": 7000000000, "sources": [ "gemma-s1", "gemma-s2" ] } ] } }, { "id": "claude-3-opus", "model": "Claude 3 Opus", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision", "structured_output" ], "tags": [ "flagship" ], "note": "Anthropic's first vision-capable flagship that topped GPT-4 across most benchmarks.", "events": [ { "type": "announcement", "date": "2024-03-04", "sources": [ "claude-3-opus-s1" ] } ], "specifications": { "language": { "context_window": 200000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-3-opus-s1", "url": "https://www.anthropic.com/news/claude-3-family", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240304143650/https://www.anthropic.com/news/claude-3-family", "retrieved": "2024-03-04" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (Mar 4, 2024) and context window (200K) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Capabilities evidenced from the primary sources: vision: “Strong vision capabilities The Claude 3 models have sophisticated vision capabilities on par with other leading models.” | structured_output: “In addition, the Claude 3 models are better at producing popular structured output in formats like JSON—making it simpler to instruct Claude for use cases like natural la”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2024-03-04", "sources": [ "claude-3-opus-s1" ] } ], "context_window": [ { "value": 200000, "sources": [ "claude-3-opus-s1" ] } ] } }, { "id": "command-r", "model": "Command R", "company": "Cohere", "family": "Command", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "agentic", "function_calling", "multilingual", "tool_use" ], "tags": [], "note": "Cohere's RAG-focused enterprise model family with 128K context and open weights.", "events": [ { "type": "announcement", "date": "2024-03-11", "sources": [ "command-r-s1", "command-r-s2" ] } ], "specifications": { "language": { "context_window": 128000, "parameter_count": 35000000000 } }, "access": { "open_weights": true, "license": "CC-BY-NC-4.0" }, "sources": [ { "id": "command-r-s1", "url": "https://cohere.com/blog/command-r", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240418190542/https://cohere.com/blog/command-r", "retrieved": "2024-04-18" }, { "id": "command-r-s2", "url": "https://huggingface.co/CohereLabs/c4ai-command-r-v01", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20250422100853/https://huggingface.co/CohereLabs/c4ai-command-r-v01", "retrieved": "2025-04-22" }, { "id": "command-r-price-20240409", "url": "https://cohere.com/pricing", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20240409111749/https://cohere.com/pricing", "retrieved": "2024-04-09" }, { "id": "command-r-price-20240913", "url": "https://cohere.com/pricing", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20240913064105/https://cohere.com/pricing", "retrieved": "2024-09-13" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (Mar 11, 2024), context window (128K) and parameter count (35B) — was found in an archived primary source, checked 2026-08-10. Modalities recorded as text-to-text. The primary source states the model is text-only. Capabilities evidenced from the primary sources: tool_use: “Model Card for Cohere Labs Command-R Model Summary Model Details Grounded Generation and RAG Capabilities: Single-Step Tool Use Capabilities ("Function Calling"” | function_calling: “Model Card for Cohere Labs Command-R Model Summary Model Details Grounded Generation and RAG Capabilities: Single-Step Tool Use Capabilities ("Function Calling"” | agentic: “Model Card for Cohere Labs Command-R Model Summary Model Details Grounded Generation and RAG Capabilities: Single-Step Tool Use Capabilities ("Function Calling"” | multilingual: “Command-R has the capability for multilingual generation evaluated in 10 languages and highly performant RAG capabilities.”." }, "evidence": { "release_date": [ { "value": "2024-03-11", "sources": [ "command-r-s1", "command-r-s2" ] } ], "context_window": [ { "value": 128000, "sources": [ "command-r-s2" ] } ], "parameter_count": [ { "value": 35000000000, "sources": [ "command-r-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 0.5, "output": 1.5 }, "currency": "USD", "observed_on": "2024-04-09", "sources": [ "command-r-price-20240409" ], "note": "Base input and output rates as the page read on this date. Fine-tuned and training rates are priced separately and are not recorded here." }, { "unit": "per_million_tokens", "rates": { "input": 0.15, "output": 0.6 }, "currency": "USD", "observed_on": "2024-09-13", "sources": [ "command-r-price-20240913" ], "note": "Base input and output rates as the page read on this date. Fine-tuned and training rates are priced separately and are not recorded here." } ] }, { "id": "llama-3", "model": "Llama 3", "company": "Meta AI", "family": "Llama", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding" ], "tags": [], "note": "8B and 70B models with dramatically stronger reasoning that powered Meta's AI assistant.", "events": [ { "type": "announcement", "date": "2024-04-18", "sources": [ "llama-3-s1", "llama-3-s2" ] } ], "specifications": { "language": { "context_window": 8192, "parameter_count": 70000000000 } }, "access": { "open_weights": true, "license": "Meta Llama 3 Community License" }, "sources": [ { "id": "llama-3-s1", "url": "https://ai.meta.com/blog/meta-llama-3/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240418155938/https://ai.meta.com/blog/meta-llama-3/", "retrieved": "2024-04-18" }, { "id": "llama-3-s2", "url": "https://huggingface.co/meta-llama/Meta-Llama-3-70B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20240418193825/https://huggingface.co/meta-llama/Meta-Llama-3-70B", "retrieved": "2024-04-18" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (April 18, 2024), context window (8,192) and parameter count (70B) — was found in an archived primary source, checked 2026-08-10. Modalities recorded as text-to-text. The primary source states the model is text-only. Capabilities evidenced from the primary sources: coding: “We also saw greatly improved capabilities like reasoning, code generation, and instruction following making Llama 3 more steerable.”." }, "evidence": { "release_date": [ { "value": "2024-04-18", "sources": [ "llama-3-s1", "llama-3-s2" ] } ], "context_window": [ { "value": 8192, "sources": [ "llama-3-s1" ] } ], "parameter_count": [ { "value": 70000000000, "sources": [ "llama-3-s1", "llama-3-s2" ] } ] } }, { "id": "phi-3-mini", "model": "Phi-3-mini", "company": "Microsoft", "family": "Phi", "classification": { "primary_type": "language", "subtype": "slm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [ "small-efficient" ], "note": "Microsoft's 3.8-billion-parameter small language model, small enough to run on a phone.", "events": [ { "type": "announcement", "date": "2024-04-23", "sources": [ "phi-3-s1", "phi-3-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 3800000000 } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "phi-3-s1", "url": "https://azure.microsoft.com/en-us/blog/introducing-phi-3-redefining-whats-possible-with-slms/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240423204115/https://azure.microsoft.com/en-us/blog/introducing-phi-3-redefining-whats-possible-with-slms/", "retrieved": "2024-04-23" }, { "id": "phi-3-s2", "url": "https://huggingface.co/microsoft/Phi-3-medium-128k-instruct", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20240521154018/https://huggingface.co/microsoft/Phi-3-medium-128k-instruct", "retrieved": "2024-05-21" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Split from the former \"phi-3\" record, which was named for the family while holding one variant's figures. Evidence: \"Phi-3-mini, a 3.8 billion parameter model. Microsoft's announcement adds that it 'is available in two context-length variants—4K and 128K tokens', so no single context window is recorded for it.\"" }, "undisclosed": [], "evidence": { "release_date": [ { "value": "2024-04-23", "sources": [ "phi-3-s1" ] } ], "parameter_count": [ { "value": 3800000000, "sources": [ "phi-3-s1" ] } ] } }, { "id": "phi-3-small", "model": "Phi-3-small", "company": "Microsoft", "family": "Phi", "classification": { "primary_type": "language", "subtype": "slm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [ "small-efficient" ], "note": "The 7-billion-parameter step in Microsoft's Phi-3 family.", "events": [ { "type": "announcement", "date": "2024-04-23", "sources": [ "phi-3-s1", "phi-3-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 7000000000 } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "phi-3-s1", "url": "https://azure.microsoft.com/en-us/blog/introducing-phi-3-redefining-whats-possible-with-slms/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240423204115/https://azure.microsoft.com/en-us/blog/introducing-phi-3-redefining-whats-possible-with-slms/", "retrieved": "2024-04-23" }, { "id": "phi-3-s2", "url": "https://huggingface.co/microsoft/Phi-3-medium-128k-instruct", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20240521154018/https://huggingface.co/microsoft/Phi-3-medium-128k-instruct", "retrieved": "2024-05-21" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Split from the former \"phi-3\" record, which was named for the family while holding one variant's figures. Evidence: \"Phi-3-small (7B) and Phi-3-medium (14B) will be available in the Azure AI model catalog and other model gardens shortly.\"" }, "undisclosed": [], "evidence": { "release_date": [ { "value": "2024-04-23", "sources": [ "phi-3-s1" ] } ], "parameter_count": [ { "value": 7000000000, "sources": [ "phi-3-s1" ] } ] } }, { "id": "phi-3-medium", "model": "Phi-3-medium", "company": "Microsoft", "family": "Phi", "classification": { "primary_type": "language", "subtype": "slm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding" ], "tags": [ "small-efficient" ], "note": "The largest Phi-3 model at 14 billion parameters, with a 128K-token context variant.", "events": [ { "type": "announcement", "date": "2024-04-23", "sources": [ "phi-3-s1", "phi-3-s2" ] } ], "specifications": { "language": { "context_window": 131072, "parameter_count": 14000000000 } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "phi-3-s1", "url": "https://azure.microsoft.com/en-us/blog/introducing-phi-3-redefining-whats-possible-with-slms/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240423204115/https://azure.microsoft.com/en-us/blog/introducing-phi-3-redefining-whats-possible-with-slms/", "retrieved": "2024-04-23" }, { "id": "phi-3-s2", "url": "https://huggingface.co/microsoft/Phi-3-medium-128k-instruct", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20240521154018/https://huggingface.co/microsoft/Phi-3-medium-128k-instruct", "retrieved": "2024-05-21" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Split from the former \"phi-3\" record, which was named for the family while holding one variant's figures. Evidence: \"Phi-3-small (7B) and Phi-3-medium (14B) will be available in the Azure AI model catalog and other model gardens shortly. The cited model card is Phi-3-medium-128k-instruct, which is where the 128K context window comes from.\" Capabilities evidenced from the primary sources: coding: “Phi-3 models are the most capable and cost-effective small language models (SLMs) available, outperforming models of the same size and next size up across a variety of la”." }, "evidence": { "release_date": [ { "value": "2024-04-23", "sources": [ "phi-3-s1" ] } ], "context_window": [ { "value": 131072, "sources": [ "phi-3-s1", "phi-3-s2" ] } ], "parameter_count": [ { "value": 14000000000, "sources": [ "phi-3-s1", "phi-3-s2" ] } ] } }, { "id": "deepseek-v2", "model": "DeepSeek-V2", "company": "DeepSeek", "family": "DeepSeek", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding" ], "tags": [], "note": "An efficient 236B MoE with 21B active parameters that made DeepSeek a global open-model force.", "events": [ { "type": "announcement", "date": "2024-05-06", "sources": [ "deepseek-v2-s1", "deepseek-v2-s2" ] } ], "specifications": { "language": { "context_window": 128000, "parameter_count": 236000000000 } }, "access": { "open_weights": true, "license": "DeepSeek License Agreement 1.0" }, "sources": [ { "id": "deepseek-v2-s1", "url": "https://arxiv.org/abs/2405.04434", "type": "technical_paper", "authority": "primary", "archived_url": "https://web.archive.org/web/20240508043415/https://arxiv.org/abs/2405.04434", "retrieved": "2024-05-08" }, { "id": "deepseek-v2-s2", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V2", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20240506170605/https://huggingface.co/deepseek-ai/DeepSeek-V2", "retrieved": "2024-05-06" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (2024-05-06), context window (128K) and parameter count (236B) — was found in an archived primary source, checked 2026-08-10. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present. Capabilities evidenced from the primary sources: coding: “As illustrated, DeepSeek-V2 demonstrates considerable proficiency in LiveCodeBench, achieving a Pass@1 score that surpasses several other sophisticated models.”." }, "evidence": { "release_date": [ { "value": "2024-05-06", "sources": [ "deepseek-v2-s2" ] } ], "context_window": [ { "value": 128000, "sources": [ "deepseek-v2-s1", "deepseek-v2-s2" ] } ], "parameter_count": [ { "value": 236000000000, "sources": [ "deepseek-v2-s1", "deepseek-v2-s2" ] } ] } }, { "id": "gpt-4o", "model": "GPT-4o", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image", "audio" ], "output": [ "text", "image", "audio" ] }, "capabilities": [ "vision", "audio" ], "tags": [ "flagship" ], "note": "OpenAI's natively multimodal real-time model with voice conversation, free in ChatGPT.", "events": [ { "type": "announcement", "date": "2024-05-13", "sources": [ "gpt-4o-s1", "gpt-4o-s2" ] } ], "specifications": { "language": { "context_window": 128000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-4o-s1", "url": "https://openai.com/index/hello-gpt-4o/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240513173656/https://openai.com/index/hello-gpt-4o/", "retrieved": "2024-05-13" }, { "id": "gpt-4o-s2", "url": "https://developers.openai.com/api/docs/models/gpt-4o", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260210080652/https://developers.openai.com/api/docs/models/gpt-4o", "retrieved": "2026-02-10" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (May 13, 2024) and context window (128,000) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Capabilities evidenced from the primary sources: vision: “GPT-4o (“o” for “omni”) is a step towards much more natural human-computer interaction—it accepts as input any combination of text, audio, and image and generates any com” | audio: “GPT-4o (“o” for “omni”) is a step towards much more natural human-computer interaction—it accepts as input any combination of text, audio, and image and generates any com”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2024-05-13", "sources": [ "gpt-4o-s1", "gpt-4o-s2" ] } ], "context_window": [ { "value": 128000, "sources": [ "gpt-4o-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 2.5, "output": 10 }, "currency": "USD", "sources": [ "gpt-4o-s2" ], "observed_on": "2026-02-10" } ] }, { "id": "qwen2", "model": "Qwen2-72B", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding", "multilingual" ], "tags": [], "note": "The largest of the five Qwen2 sizes, trained on data in 27 languages beyond English and Chinese.", "events": [ { "type": "announcement", "date": "2024-06-07", "sources": [ "qwen2-s1", "qwen2-s2" ] } ], "specifications": { "language": { "context_window": 131072, "parameter_count": 72710000000 } }, "access": { "open_weights": true, "license": "Tongyi Qianwen License Agreement" }, "sources": [ { "id": "qwen2-s1", "url": "https://qwenlm.github.io/blog/qwen2/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240606203728/https://qwenlm.github.io/blog/qwen2/", "retrieved": "2024-06-06" }, { "id": "qwen2-s2", "url": "https://huggingface.co/Qwen/Qwen2-72B-Instruct", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20240606201220/https://huggingface.co/Qwen/Qwen2-72B-Instruct", "retrieved": "2024-06-06" } ], "provenance": { "status": "partially_verified", "confidence": 80, "reason": "Cited to the lab's own announcement and the official model card. Not verified because the parameter count could not be traced: every archived snapshot of those sources predates the release — the newest is 2024-06-06, captured before the page announced it — so the archived copy does not contain the figure. A capture from on or after 2024-06-07 would settle it." }, "evidence": { "release_date": [ { "value": "2024-06-07", "sources": [ "qwen2-s1" ] } ], "context_window": [ { "value": 131072, "sources": [ "qwen2-s1", "qwen2-s2" ] } ] } }, { "id": "nemotron-4-340b", "model": "Nemotron-4 340B", "company": "NVIDIA", "family": "Nemotron", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [], "tags": [], "note": "Shipped as base, instruct and reward models under a permissive open licence, built to generate synthetic training data.", "events": [ { "type": "announcement", "date": "2024-06-14", "sources": [ "nemotron-4-340b-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 340000000000 } }, "access": { "open_weights": true, "license": "NVIDIA Open Model License" }, "sources": [ { "id": "nemotron-4-340b-s1", "url": "https://blogs.nvidia.com/blog/nemotron-4-synthetic-data-generation-llm-training/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240614155430/https://blogs.nvidia.com/blog/nemotron-4-synthetic-data-generation-llm-training/", "retrieved": "2024-06-14" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2024-06-14", "sources": [ "nemotron-4-340b-s1" ] } ], "parameter_count": [ { "value": 340000000000, "sources": [ "nemotron-4-340b-s1" ] } ] }, "undisclosed": [ "context_window" ] }, { "id": "claude-3-5-sonnet", "model": "Claude 3.5 Sonnet", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision" ], "tags": [ "flagship" ], "note": "The mid-tier model that became 2024's favorite for coding and agentic workflows.", "events": [ { "type": "announcement", "date": "2024-06-21", "sources": [ "claude-3-5-sonnet-s1" ] } ], "specifications": { "language": { "context_window": 200000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-3-5-sonnet-s1", "url": "https://www.anthropic.com/news/claude-3-5-sonnet", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240620140357/https://www.anthropic.com/news/claude-3-5-sonnet", "retrieved": "2024-06-20" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2024-06-21", "sources": [ "claude-3-5-sonnet-s1" ] } ], "context_window": [ { "value": 200000, "sources": [ "claude-3-5-sonnet-s1" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 3, "output": 15 }, "currency": "USD", "sources": [ "claude-3-5-sonnet-s1" ], "observed_on": "2024-06-20" } ] }, { "id": "llama-3-1", "model": "Llama 3.1", "company": "Meta AI", "family": "Llama", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "multilingual", "tool_use" ], "tags": [], "note": "The 405B open flagship with 128K context, the first open model at frontier scale.", "events": [ { "type": "announcement", "date": "2024-07-23", "sources": [ "llama-3-1-s1", "llama-3-1-s2" ] } ], "specifications": { "language": { "context_window": 128000, "parameter_count": 405000000000 } }, "access": { "open_weights": true, "license": "Llama 3.1 Community License" }, "sources": [ { "id": "llama-3-1-s1", "url": "https://ai.meta.com/blog/meta-llama-3-1/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240723153909/https://ai.meta.com/blog/meta-llama-3-1/", "retrieved": "2024-07-23" }, { "id": "llama-3-1-s2", "url": "https://huggingface.co/meta-llama/Llama-3.1-405B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20240927110330/https://huggingface.co/meta-llama/Llama-3.1-405B", "retrieved": "2024-09-27" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (July 23, 2024), context window (128K) and parameter count (405B) — was found in an archived primary source, checked 2026-08-10. Modalities recorded as text-to-text. The primary source states the model is text-only. Capabilities evidenced from the primary sources: multilingual: “As part of our ongoing efforts to develop AI responsibly beyond the model layer and helping others to do the same, we’re releasing a full reference system that includes s”." }, "evidence": { "release_date": [ { "value": "2024-07-23", "sources": [ "llama-3-1-s1", "llama-3-1-s2" ] } ], "context_window": [ { "value": 128000, "sources": [ "llama-3-1-s1", "llama-3-1-s2" ] } ], "parameter_count": [ { "value": 405000000000, "sources": [ "llama-3-1-s1", "llama-3-1-s2" ] } ] } }, { "id": "mistral-large-2", "model": "Mistral Large 2", "company": "Mistral AI", "family": "Mistral", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding", "function_calling", "multilingual", "tool_use" ], "tags": [ "flagship" ], "note": "Mistral's 123-billion-parameter model rivaling closed frontier models in multilingual tasks.", "events": [ { "type": "announcement", "date": "2024-07-24", "sources": [ "mistral-large-2-s1", "mistral-large-2-s2" ] } ], "specifications": { "language": { "context_window": 128000, "parameter_count": 123000000000 } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mistral-large-2-s1", "url": "https://mistral.ai/news/mistral-large-2407/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240724152547/https://mistral.ai/news/mistral-large-2407/", "retrieved": "2024-07-24" }, { "id": "mistral-large-2-s2", "url": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2407", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20240724171028/https://huggingface.co/mistralai/Mistral-Large-Instruct-2407", "retrieved": "2024-07-24" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (July 24, 2024), context window (128k) and parameter count (123 billion) — was found in an archived primary source, checked 2026-08-10. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present. Capabilities evidenced from the primary sources: coding: “Compared to its predecessor, Mistral Large 2 is significantly more capable in code generation, mathematics, and reasoning.” | multilingual: “While the majority of models are English-centric, the new Mistral Large 2 was trained on a large proportion of multilingual data.” | tool_use: “Performance on Multilingual MMLU (measured on the base pretrained model) Tool Use & Function Calling Mistral Large 2 is equipped with enhanced function calling and retrie” | function_calling: “Performance on Multilingual MMLU (measured on the base pretrained model) Tool Use & Function Calling Mistral Large 2 is equipped with enhanced function calling and retrie”." }, "evidence": { "release_date": [ { "value": "2024-07-24", "sources": [ "mistral-large-2-s1", "mistral-large-2-s2" ] } ], "context_window": [ { "value": 128000, "sources": [ "mistral-large-2-s1", "mistral-large-2-s2" ] } ], "parameter_count": [ { "value": 123000000000, "sources": [ "mistral-large-2-s1", "mistral-large-2-s2" ] } ] } }, { "id": "jamba-1-5", "model": "Jamba 1.5", "company": "AI21 Labs", "family": "Jamba", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "agentic", "tool_use" ], "tags": [], "note": "AI21's hybrid SSM-Transformer models with a 256K context window for long-document work.", "events": [ { "type": "announcement", "date": "2024-08-22", "sources": [ "jamba-1-5-s1", "jamba-1-5-s2" ] } ], "specifications": { "language": { "context_window": 256000, "parameter_count": 398000000000 } }, "access": { "open_weights": true, "license": "Jamba Open Model License" }, "sources": [ { "id": "jamba-1-5-s1", "url": "https://www.ai21.com/blog/announcing-jamba-model-family/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240822135946/https://www.ai21.com/blog/announcing-jamba-model-family", "retrieved": "2024-08-22" }, { "id": "jamba-1-5-s2", "url": "https://huggingface.co/ai21labs/AI21-Jamba-1.5-Large", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20240822135916/https://huggingface.co/ai21labs/AI21-Jamba-1.5-Large", "retrieved": "2024-08-22" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (August 22, 2024), context window (256K) and parameter count (398B) — was found in an archived primary source, checked 2026-08-10. Capabilities evidenced from the primary sources: tool_use: “5-Large\" , use_mamba_kernels= False ) Model features Tool use with Jamba Jamba 1.”." }, "evidence": { "release_date": [ { "value": "2024-08-22", "sources": [ "jamba-1-5-s1", "jamba-1-5-s2" ] } ], "context_window": [ { "value": 256000, "sources": [ "jamba-1-5-s1", "jamba-1-5-s2" ] } ], "parameter_count": [ { "value": 398000000000, "sources": [ "jamba-1-5-s2" ] } ] } }, { "id": "deepseek-v2-5", "model": "DeepSeek-V2.5", "company": "DeepSeek", "family": "DeepSeek", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding", "function_calling", "structured_output" ], "tags": [], "note": "An all-in-one merge of DeepSeek-V2-Chat and DeepSeek-Coder-V2 that combines general conversational and code capabilities.", "events": [ { "type": "announcement", "date": "2024-09-05", "sources": [ "deepseek-v2-5-s1", "deepseek-v2-5-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 236000000000 } }, "access": { "open_weights": true, "license": "DeepSeek License Agreement 1.0" }, "sources": [ { "id": "deepseek-v2-5-s1", "url": "https://api-docs.deepseek.com/news/news0905/", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20241112035822/http://api-docs.deepseek.com/news/news0905/", "retrieved": "2024-11-12" }, { "id": "deepseek-v2-5-s2", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V2.5", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20240906133246/https://huggingface.co/deepseek-ai/DeepSeek-V2.5", "retrieved": "2024-09-06" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to official documentation and the official model card, and every value it asserts was found there. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present. Capabilities evidenced from the primary sources: coding: “5: A New Open-Source Model Combining General and Coding Capabilities | DeepSeek API Docs Skip to main content DeepSeek API Docs English English 中文（中国） DeepSeek Platform Q” | function_calling: “5 Release 2024/09/05 Context Caching is Available 2024/08/02 New API Features 2024/07/25 API Reference API Guides Multi-round Conversation Chat Prefix Completion (Beta) F” | structured_output: “5 Release 2024/09/05 Context Caching is Available 2024/08/02 New API Features 2024/07/25 API Reference API Guides Multi-round Conversation Chat Prefix Completion (Beta) F”." }, "evidence": { "release_date": [ { "value": "2024-09-05", "sources": [ "deepseek-v2-5-s2" ] } ], "parameter_count": [ { "value": 236000000000, "sources": [ "deepseek-v2-5-s2" ] } ] }, "undisclosed": [ "context_window" ] }, { "id": "o1", "model": "o1", "company": "OpenAI", "family": "o-series", "classification": { "primary_type": "language", "subtype": "reasoning" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "vision" ], "tags": [], "note": "OpenAI's first reasoning model that thinks before answering, launching the reasoning-model era.", "events": [ { "type": "announcement", "date": "2024-09-12", "sources": [ "o1-s1", "o1-s2" ] } ], "specifications": { "language": { "context_window": 200000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "o1-s1", "url": "https://openai.com/index/learning-to-reason-with-llms/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240912171926/https://openai.com/index/learning-to-reason-with-llms/", "retrieved": "2024-09-12" }, { "id": "o1-s2", "url": "https://developers.openai.com/api/docs/models/o1", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260320163509/https://developers.openai.com/api/docs/models/o1", "retrieved": "2026-03-20" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (September 12, 2024) and context window (200,000) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Modalities taken from the documentation's own declaration: \"Input Text, image Output Text\". Capabilities evidenced from the primary sources: vision: “Modalities Text Input and output Image Input only Audio Not supported Video Not supported” | reasoning: “Reasoning token support”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2024-09-12", "sources": [ "o1-s1", "o1-s2" ] } ], "context_window": [ { "value": 200000, "sources": [ "o1-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 15, "output": 60 }, "currency": "USD", "sources": [ "o1-s2" ], "observed_on": "2026-03-20" } ], "benchmarks": [ { "name": "MMMU", "score": 78.2, "evaluation_type": "vendor_reported", "reported_on": "2024-09-12", "sources": [ "o1-s1" ] } ] }, { "id": "qwen2-5", "model": "Qwen2.5-72B", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "multilingual" ], "tags": [], "note": "The largest open Qwen2.5 model, released alongside the Qwen2.5-Coder and Qwen2.5-Math lines.", "events": [ { "type": "announcement", "date": "2024-09-19", "sources": [ "qwen2-5-s1", "qwen2-5-s2" ] } ], "specifications": { "language": { "context_window": 131072, "parameter_count": 72000000000 } }, "access": { "open_weights": true, "license": "Qwen License Agreement" }, "sources": [ { "id": "qwen2-5-s1", "url": "https://qwenlm.github.io/blog/qwen2.5/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240918201754/https://qwenlm.github.io/blog/qwen2.5/", "retrieved": "2024-09-18" }, { "id": "qwen2-5-s2", "url": "https://huggingface.co/Qwen/Qwen2.5-72B-Instruct", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20240920075202/https://huggingface.co/Qwen/Qwen2.5-72B-Instruct", "retrieved": "2024-09-20" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Renamed from the series to the shipped flagship, which Alibaba describes as \"a 72B-parameter dense decoder-only language model\". The 72B and 3B variants are excluded from the series-wide Apache 2.0 licence." }, "evidence": { "release_date": [ { "value": "2024-09-19", "sources": [ "qwen2-5-s1", "qwen2-5-s2" ] } ], "context_window": [ { "value": 131072, "sources": [ "qwen2-5-s2" ] } ], "parameter_count": [ { "value": 72000000000, "sources": [ "qwen2-5-s1", "qwen2-5-s2" ] } ] } }, { "id": "llama-3-2", "model": "Llama 3.2", "company": "Meta AI", "family": "Llama", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision" ], "tags": [], "note": "The first vision-capable Llama plus 1B and 3B models for edge devices.", "events": [ { "type": "announcement", "date": "2024-09-25", "sources": [ "llama-3-2-s1", "llama-3-2-s2" ] } ], "specifications": { "language": { "context_window": 128000, "parameter_count": 90000000000 } }, "access": { "open_weights": true, "license": "Llama 3.2 Community License" }, "sources": [ { "id": "llama-3-2-s1", "url": "https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20240925173233/https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/", "retrieved": "2024-09-25" }, { "id": "llama-3-2-s2", "url": "https://huggingface.co/meta-llama/Llama-3.2-90B-Vision-Instruct", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20240925174046/https://huggingface.co/meta-llama/Llama-3.2-90B-Vision-Instruct", "retrieved": "2024-09-25" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (2024-09-25), context window (128k) and parameter count (90B) — was found in an archived primary source, checked 2026-08-10. Capabilities evidenced from the primary sources: vision: “To support image recognition tasks, the Llama 3.”." }, "evidence": { "release_date": [ { "value": "2024-09-25", "sources": [ "llama-3-2-s2" ] } ], "context_window": [ { "value": 128000, "sources": [ "llama-3-2-s2" ] } ], "parameter_count": [ { "value": 90000000000, "sources": [ "llama-3-2-s2" ] } ] } }, { "id": "nova-micro", "model": "Nova Micro", "company": "Amazon", "family": "Nova", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding" ], "tags": [], "note": "Amazon's lowest-latency Nova model, text in and text out.", "events": [ { "type": "announcement", "date": "2024-12-03", "sources": [ "nova-s1", "nova-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "nova-s1", "url": "https://www.aboutamazon.com/news/aws/amazon-nova-artificial-intelligence-bedrock-aws", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20241203175425/https://www.aboutamazon.com/news/aws/amazon-nova-artificial-intelligence-bedrock-aws", "retrieved": "2024-12-03" }, { "id": "nova-s2", "url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20241203181439/https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/", "retrieved": "2024-12-03" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Split from the former single \"Nova\" family record. Every value is quoted from an archived AWS announcement: \"Amazon Nova Micro – A text-only model that delivers the lowest latency responses in the Amazon Nova family of models at a very low cost.\" Capabilities evidenced from the primary sources: coding: “With a context length of 128K tokens and optimized for speed and cost, Amazon Nova Micro excels at tasks such as text summarization, translation, content classification, ”." }, "undisclosed": [ "context_window", "parameter_count" ], "evidence": { "release_date": [ { "value": "2024-12-03", "sources": [ "nova-s1" ] } ] } }, { "id": "nova-lite", "model": "Nova Lite", "company": "Amazon", "family": "Nova", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": [ "vision", "video" ], "tags": [], "note": "Amazon's low-cost multimodal Nova model, reading images and video to produce text.", "events": [ { "type": "announcement", "date": "2024-12-03", "sources": [ "nova-s1", "nova-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "nova-s1", "url": "https://www.aboutamazon.com/news/aws/amazon-nova-artificial-intelligence-bedrock-aws", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20241203175425/https://www.aboutamazon.com/news/aws/amazon-nova-artificial-intelligence-bedrock-aws", "retrieved": "2024-12-03" }, { "id": "nova-s2", "url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20241203181439/https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/", "retrieved": "2024-12-03" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Split from the former single \"Nova\" family record. Every value is quoted from an archived AWS announcement: \"Amazon Nova Lite – A very low-cost multimodal model that is lightning fast for processing image, video, and text inputs to generate text output.\" Capabilities evidenced from the primary sources: video: “You can build on Amazon Nova to analyze complex documents and videos, understand charts and diagrams, generate engaging video content, and build sophisticated AI agents, ” | vision: “Whether you’re developing document processing applications that need to process images and text, creating marketing content at scale, or building AI assistants that can u”." }, "undisclosed": [ "context_window", "parameter_count" ], "evidence": { "release_date": [ { "value": "2024-12-03", "sources": [ "nova-s1" ] } ] } }, { "id": "nova-pro", "model": "Nova Pro", "company": "Amazon", "family": "Nova", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": [ "vision", "video", "agentic", "tool_use" ], "tags": [], "note": "Amazon's most capable shipped Nova understanding model, with a 300K-token context window.", "events": [ { "type": "announcement", "date": "2024-12-03", "sources": [ "nova-s1", "nova-s2" ] } ], "specifications": { "language": { "context_window": 300000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "nova-s1", "url": "https://www.aboutamazon.com/news/aws/amazon-nova-artificial-intelligence-bedrock-aws", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20241203175425/https://www.aboutamazon.com/news/aws/amazon-nova-artificial-intelligence-bedrock-aws", "retrieved": "2024-12-03" }, { "id": "nova-s2", "url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20241203181439/https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/", "retrieved": "2024-12-03" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Split from the former single \"Nova\" family record. Every value is quoted from an archived AWS announcement: \"Amazon Nova Pro is capable of processing up to 300K input tokens and sets new standards in multimodal intelligence and agentic workflows.\" Capabilities evidenced from the primary sources: agentic: ““Our new Amazon Nova models are intended to help with these challenges for internal and external builders, and provide compelling intelligence and content generation whil” | vision: “Amazon Nova Reel transforms a single image input into a brief video with the prompt: dolly forward .” | video: “Read more Video understanding To put the video understanding capabilities of Amazon Nova Pro to the test, we asked the model to review and describe a 12-second silent vid”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2024-12-03", "sources": [ "nova-s1" ] } ], "context_window": [ { "value": 300000, "sources": [ "nova-s2" ] } ] } }, { "id": "nova-canvas", "model": "Nova Canvas", "company": "Amazon", "family": "Nova Canvas", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": { "input": [ "text", "image" ], "output": [ "image" ] }, "capabilities": [ "image_generation" ], "tags": [], "note": "Amazon's image generation model, with inpainting, outpainting and background editing.", "events": [ { "type": "announcement", "date": "2024-12-03", "sources": [ "nova-s1", "nova-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "nova-s1", "url": "https://www.aboutamazon.com/news/aws/amazon-nova-artificial-intelligence-bedrock-aws", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20241203175425/https://www.aboutamazon.com/news/aws/amazon-nova-artificial-intelligence-bedrock-aws", "retrieved": "2024-12-03" }, { "id": "nova-s2", "url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20241203181439/https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/", "retrieved": "2024-12-03" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Split from the former single \"Nova\" family record. Every value is quoted from an archived AWS announcement: \"Amazon Nova Canvas – A state-of-the-art image generation model producing studio-quality images with precise control over style and content, including rich editing features such as inpainting, outpainting, and background removal.\" Capabilities evidenced from the primary sources: image_generation: “Amazon Nova Canvas , a state-of-the-art image generation model.”." }, "evidence": { "release_date": [ { "value": "2024-12-03", "sources": [ "nova-s1" ] } ] } }, { "id": "nova-reel", "model": "Nova Reel", "company": "Amazon", "family": "Nova Reel", "classification": { "primary_type": "video_generation", "subtype": null }, "modalities": { "input": [ "text", "image" ], "output": [ "video" ] }, "capabilities": [ "video_generation" ], "tags": [], "note": "Amazon's video generation model, producing short videos from text prompts and images.", "events": [ { "type": "announcement", "date": "2024-12-03", "sources": [ "nova-s1", "nova-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "nova-s1", "url": "https://www.aboutamazon.com/news/aws/amazon-nova-artificial-intelligence-bedrock-aws", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20241203175425/https://www.aboutamazon.com/news/aws/amazon-nova-artificial-intelligence-bedrock-aws", "retrieved": "2024-12-03" }, { "id": "nova-s2", "url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20241203181439/https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/", "retrieved": "2024-12-03" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Split from the former single \"Nova\" family record. Every value is quoted from an archived AWS announcement: \"Amazon Nova Reel – A state-of-the-art video generation model. With Amazon Nova Reel, you can produce short videos through text prompts and images.\" Capabilities evidenced from the primary sources: video_generation: “Amazon Nova Reel , a state-of-the-art video generation model.”." }, "evidence": { "release_date": [ { "value": "2024-12-03", "sources": [ "nova-s1" ] } ] } }, { "id": "gemini-2-flash", "model": "Gemini 2.0 Flash", "company": "Google DeepMind", "family": "Gemini", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "audio", "image", "video", "text" ], "output": [ "text" ] }, "capabilities": [ "agentic", "function_calling", "structured_output" ], "tags": [ "flagship" ], "note": "Google's agentic-era model with native tool use and real-time multimodal understanding.", "events": [ { "type": "announcement", "date": "2024-12-11", "sources": [ "gemini-2-flash-s1", "gemini-2-flash-s2" ] } ], "specifications": { "language": { "context_window": 1048576, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-2-flash-s1", "url": "https://blog.google/technology/google-deepmind/google-gemini-ai-update-december-2024/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20241211154129/https://blog.google/technology/google-deepmind/google-gemini-ai-update-december-2024/", "retrieved": "2024-12-11" }, { "id": "gemini-2-flash-s2", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.0-flash", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260303034822/https://ai.google.dev/gemini-api/docs/models/gemini-2.0-flash", "retrieved": "2026-03-03" }, { "id": "gemini-2-flash-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.0-flash?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260303034822/https://ai.google.dev/gemini-api/docs/models/gemini-2.0-flash?hl=en", "retrieved": "2026-03-03" }, { "id": "gemini-2-flash-price-20251103", "url": "https://ai.google.dev/gemini-api/docs/pricing", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20251103074340/https://ai.google.dev/gemini-api/docs/pricing", "retrieved": "2025-11-03" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (Dec 11, 2024) and context window (1,048,576) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Modalities taken from the documentation's own declaration: \"Inputs Audio, images, video, and text Output Text\". Capabilities evidenced from the primary sources: agentic: “0: our new AI model for the agentic era Dec 11, 2024 · [[read-time]] min read Share Twitter Facebook LinkedIn Mail Copy link Sundar Pichai CEO of Google and Alphabet Demi”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2024-12-11", "sources": [ "gemini-2-flash-s1" ] } ], "context_window": [ { "value": 1048576, "sources": [ "gemini-2-flash-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 0.1, "output": 0.4 }, "currency": "USD", "observed_on": "2025-11-03", "sources": [ "gemini-2-flash-price-20251103" ], "note": "Paid-tier input and output as the page read on this date. The free tier and context-caching rates are priced separately and are not recorded here." } ] }, { "id": "deepseek-v3", "model": "DeepSeek-V3", "company": "DeepSeek", "family": "DeepSeek", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "A 671B MoE open model whose near-frontier performance at minimal cost stunned the industry.", "events": [ { "type": "announcement", "date": "2024-12-26", "sources": [ "deepseek-v3-s3" ] } ], "specifications": { "language": { "context_window": 128000, "parameter_count": 671000000000 } }, "access": { "open_weights": true, "license": "DeepSeek License Agreement 1.0" }, "sources": [ { "id": "deepseek-v3-s1", "url": "https://arxiv.org/abs/2412.19437", "type": "technical_paper", "authority": "primary", "archived_url": "https://web.archive.org/web/20241230045532/https://arxiv.org/abs/2412.19437", "retrieved": "2024-12-30" }, { "id": "deepseek-v3-s2", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20241226081224/https://huggingface.co/deepseek-ai/DeepSeek-V3", "retrieved": "2024-12-26" }, { "id": "deepseek-v3-s3", "url": "https://api-docs.deepseek.com/news/news1226/", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260811022652/https://api-docs.deepseek.com/news/news1226/", "retrieved": "2026-08-10" }, { "id": "deepseek-v3-price-20250306", "url": "https://api-docs.deepseek.com/quick_start/pricing/", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20250306121853/https://api-docs.deepseek.com/quick_start/pricing/", "retrieved": "2025-03-06" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the technical paper and the official model card. Not verified because the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present." }, "evidence": { "release_date": [ { "value": "2024-12-26", "sources": [ "deepseek-v3-s2", "deepseek-v3-s3", "deepseek-v3-price-20250306" ] } ], "context_window": [ { "value": 128000, "sources": [ "deepseek-v3-s2" ] } ], "parameter_count": [ { "value": 671000000000, "sources": [ "deepseek-v3-s1", "deepseek-v3-s2", "deepseek-v3-s3" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 0.27, "output": 1.1 }, "currency": "USD", "observed_on": "2025-03-06", "sources": [ "deepseek-v3-price-20250306" ], "note": "Standard-hours cache-miss input and output as the page read on this date. Cache-hit input and any off-peak discount are priced separately and are not recorded here." } ] }, { "id": "deepseek-r1", "model": "DeepSeek-R1", "company": "DeepSeek", "family": "DeepSeek", "classification": { "primary_type": "language", "subtype": "reasoning" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "reasoning" ], "tags": [], "note": "The open reasoning model that matched o1-class performance and triggered a global tech selloff.", "events": [ { "type": "announcement", "date": "2025-01-20", "sources": [ "deepseek-r1-s1", "deepseek-r1-s2" ] } ], "specifications": { "language": { "context_window": 128000, "parameter_count": 671000000000 } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "deepseek-r1-s1", "url": "https://arxiv.org/abs/2501.12948", "type": "technical_paper", "authority": "primary", "archived_url": "https://web.archive.org/web/20250123030008/https://arxiv.org/abs/2501.12948", "retrieved": "2025-01-23" }, { "id": "deepseek-r1-s2", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20250120073421/https://huggingface.co/deepseek-ai/DeepSeek-R1", "retrieved": "2025-01-20" }, { "id": "deepseek-r1-price-20250306", "url": "https://api-docs.deepseek.com/quick_start/pricing/", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20250306121853/https://api-docs.deepseek.com/quick_start/pricing/", "retrieved": "2025-03-06" } ], "provenance": { "status": "partially_verified", "confidence": 75, "reason": "Cited to the technical paper and the official model card. Not verified because the context window and the parameter count could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present. Capabilities evidenced from the primary sources: reasoning: “View a PDF of the paper titled DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning, by DeepSeek-AI and 199 other authors View PDF HTML (exp”." }, "evidence": { "release_date": [ { "value": "2025-01-20", "sources": [ "deepseek-r1-s2", "deepseek-r1-price-20250306" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 0.55, "output": 2.19 }, "currency": "USD", "observed_on": "2025-03-06", "sources": [ "deepseek-r1-price-20250306" ], "note": "Standard-hours cache-miss input and output as the page read on this date. Cache-hit input and any off-peak discount are priced separately and are not recorded here." } ] }, { "id": "mistral-small-3", "model": "Mistral Small 3", "company": "Mistral AI", "family": "Mistral", "classification": { "primary_type": "language", "subtype": "slm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "agentic", "function_calling", "multilingual" ], "tags": [ "small-efficient" ], "note": "A fast 24B Apache-licensed model designed for on-device and low-latency deployment.", "events": [ { "type": "announcement", "date": "2025-01-30", "sources": [ "mistral-small-3-s1", "mistral-small-3-s2" ] } ], "specifications": { "language": { "context_window": 32768, "parameter_count": 24000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "mistral-small-3-s1", "url": "https://mistral.ai/news/mistral-small-3/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250130143452/https://mistral.ai/news/mistral-small-3/", "retrieved": "2025-01-30" }, { "id": "mistral-small-3-s2", "url": "https://huggingface.co/mistralai/Mistral-Small-24B-Instruct-2501", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20250130151122/https://huggingface.co/mistralai/Mistral-Small-24B-Instruct-2501", "retrieved": "2025-01-30" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement and the official model card. Not verified because the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present. Capabilities evidenced from the primary sources: function_calling: “Low-latency function calling: Mistral Small 3 is able to handle rapid function execution when used as part of automated or agentic workflows.” | agentic: “Low-latency function calling: Mistral Small 3 is able to handle rapid function execution when used as part of automated or agentic workflows.” | multilingual: “25k Safetensors 10 languages vllm mistral License: apache-2.”." }, "evidence": { "release_date": [ { "value": "2025-01-30", "sources": [ "mistral-small-3-s1", "mistral-small-3-s2" ] } ], "context_window": [ { "value": 32768, "sources": [ "mistral-small-3-s2" ] } ], "parameter_count": [ { "value": 24000000000, "sources": [ "mistral-small-3-s1", "mistral-small-3-s2" ] } ] } }, { "id": "grok-3", "model": "Grok 3", "company": "xAI", "family": "Grok", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "vision", "coding" ], "tags": [ "flagship" ], "note": "Trained on 200K GPUs with DeepSearch integration, briefly topping public leaderboards.", "events": [ { "type": "announcement", "date": "2025-02-19", "sources": [ "grok-3-s1", "grok-3-s2" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "grok-3-s1", "url": "https://x.ai/news/grok-3", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250311172142/https://x.ai/news/grok-3", "retrieved": "2025-03-11" }, { "id": "grok-3-s2", "url": "https://artificialanalysis.ai/models/grok-3", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20250218213019/https://artificialanalysis.ai/models/grok-3", "retrieved": "2025-02-18" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2025-02-19", "sources": [ "grok-3-s1" ] } ], "context_window": [ { "value": 1000000, "sources": [ "grok-3-s1" ] } ] }, "benchmarks": [ { "name": "LiveCodeBench", "score": 79.4, "evaluation_type": "vendor_reported", "reported_on": "2025-02-17", "sources": [ "grok-3-s1" ] }, { "name": "AIME", "score": 95.8, "evaluation_type": "vendor_reported", "reported_on": "2025-02-17", "sources": [ "grok-3-s1" ] } ] }, { "id": "claude-3-7-sonnet", "model": "Claude 3.7 Sonnet", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "vision" ], "tags": [ "flagship" ], "note": "The first hybrid reasoning model with a user-controlled extended thinking mode.", "events": [ { "type": "announcement", "date": "2025-02-24", "sources": [ "claude-3-7-sonnet-s1", "claude-3-7-sonnet-s2" ] } ], "specifications": { "language": { "context_window": 200000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-3-7-sonnet-s1", "url": "https://www.anthropic.com/news/claude-3-7-sonnet", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250224183042/https://www.anthropic.com/news/claude-3-7-sonnet", "retrieved": "2025-02-24" }, { "id": "claude-3-7-sonnet-s2", "url": "https://artificialanalysis.ai/models/claude-3-7-sonnet", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20250312184830/https://artificialanalysis.ai/models/claude-3-7-sonnet", "retrieved": "2025-03-12" }, { "id": "claude-3-7-sonnet-s3", "url": "https://docs.anthropic.com/en/docs/about-claude/models/overview", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20250519172951/https://docs.anthropic.com/en/docs/about-claude/models/overview", "retrieved": "2025-05-19" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement. Not verified because the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities read from Anthropic's model documentation as it stood on 2025-05-19, which lists this model as: “Claude 3.7 Sonnet Our most intelligent model Text and image input Text output 200k context window Extended thinking.” The current documentation no longer lists it, so the contemporaneous snapshot is cited rather than the live page. Capabilities evidenced from the primary sources: reasoning: “Extended thinking mode is available on all surfaces except the free Claude tier.” | vision: “5 Haiku Our fastest model Text and image input Text output 200k context window Claude 3.”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2025-02-24", "sources": [ "claude-3-7-sonnet-s1" ] } ], "context_window": [ { "value": 200000, "sources": [ "claude-3-7-sonnet-s3" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 3, "output": 15 }, "currency": "USD", "sources": [ "claude-3-7-sonnet-s1" ], "observed_on": "2025-02-24" } ] }, { "id": "gpt-4-5", "model": "GPT-4.5", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision" ], "tags": [ "flagship" ], "note": "OpenAI's final GPT-4-generation model, strongest at natural conversation and world knowledge.", "events": [ { "type": "announcement", "date": "2025-02-27", "sources": [ "gpt-4-5-s1", "gpt-4-5-s2" ] } ], "specifications": { "language": { "context_window": 128000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-4-5-s1", "url": "https://openai.com/index/introducing-gpt-4-5/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250228042933/https://openai.com/index/introducing-gpt-4-5/", "retrieved": "2025-02-28" }, { "id": "gpt-4-5-s2", "url": "https://developers.openai.com/api/docs/models/gpt-4.5-preview", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260211085909/https://developers.openai.com/api/docs/models/gpt-4.5-preview", "retrieved": "2026-02-11" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (February 27, 2025) and context window (128,000) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Capabilities evidenced from the primary sources: vision: “Modalities Text Input and output Image Input only Audio Not supported Video Not supported”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2025-02-27", "sources": [ "gpt-4-5-s1", "gpt-4-5-s2" ] } ], "context_window": [ { "value": 128000, "sources": [ "gpt-4-5-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 75, "output": 150 }, "currency": "USD", "sources": [ "gpt-4-5-s2" ], "observed_on": "2026-02-11" } ] }, { "id": "qwq-32b", "model": "QwQ-32B", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "reasoning" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "reasoning" ], "tags": [], "note": "A 32B open reasoning model trained with reinforcement learning, released under Apache 2.0.", "events": [ { "type": "announcement", "date": "2025-03-06", "sources": [ "qwq-32b-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 32000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "qwq-32b-s1", "url": "https://qwenlm.github.io/blog/qwq-32b/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250305205353/https://qwenlm.github.io/blog/qwq-32b/", "retrieved": "2025-03-05" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present." }, "evidence": { "release_date": [ { "value": "2025-03-06", "sources": [ "qwq-32b-s1" ] } ], "parameter_count": [ { "value": 32000000000, "sources": [ "qwq-32b-s1" ] } ] }, "undisclosed": [ "context_window" ] }, { "id": "gemma-3", "model": "Gemma 3", "company": "Google DeepMind", "family": "Gemma", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision", "function_calling", "structured_output" ], "tags": [], "note": "Google's open family adding vision, 128K context, and support for 140 languages.", "events": [ { "type": "announcement", "date": "2025-03-12", "sources": [ "gemma-3-s1", "gemma-3-s2", "gemma-3-s3" ] } ], "specifications": { "language": { "context_window": 131072, "parameter_count": 27000000000 } }, "access": { "open_weights": true, "license": "Gemma Terms of Use" }, "sources": [ { "id": "gemma-3-s1", "url": "https://blog.google/technology/developers/gemma-3/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250312085706/https://blog.google/technology/developers/gemma-3/", "retrieved": "2025-03-12" }, { "id": "gemma-3-s2", "url": "https://huggingface.co/google/gemma-3-27b-it", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20250312064154/https://huggingface.co/google/gemma-3-27b-it", "retrieved": "2025-03-12" }, { "id": "gemma-3-s3", "url": "https://ai.google.dev/gemma/terms", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250313154706/https://ai.google.dev/gemma/terms", "retrieved": "2025-03-13" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement and the official model card. Not verified because the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Capabilities evidenced from the primary sources: vision: “Gemma 3 supports over 140 languages and offers advanced text and visual reasoning capabilities.” | function_calling: “Gemma 3 includes a 128k-token context window and supports function calling for complex tasks.” | structured_output: “Create AI-driven workflows using function calling: Gemma 3 supports function calling and structured output to help you automate tasks and build agentic experiences.”." }, "evidence": { "release_date": [ { "value": "2025-03-12", "sources": [ "gemma-3-s1", "gemma-3-s2" ] } ], "context_window": [ { "value": 131072, "sources": [ "gemma-3-s1", "gemma-3-s2", "gemma-3-s3" ] } ], "parameter_count": [ { "value": 27000000000, "sources": [ "gemma-3-s1", "gemma-3-s2" ] } ] } }, { "id": "command-a", "model": "Command A", "company": "Cohere", "family": "Command", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "agentic", "multilingual", "tool_use" ], "tags": [ "flagship" ], "note": "Cohere's 111B dense flagship for enterprise RAG and agent workloads with 256K context.", "events": [ { "type": "announcement", "date": "2025-03-13", "sources": [ "command-a-s1", "command-a-s2" ] } ], "specifications": { "language": { "context_window": 256000, "parameter_count": 111000000000 } }, "access": { "open_weights": true, "license": "CC-BY-NC-4.0" }, "sources": [ { "id": "command-a-s1", "url": "https://cohere.com/blog/command-a", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250313130452/https://cohere.com/blog/command-a", "retrieved": "2025-03-13" }, { "id": "command-a-s2", "url": "https://huggingface.co/CohereLabs/c4ai-command-a-03-2025", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20250416154400/https://huggingface.co/CohereLabs/c4ai-command-a-03-2025", "retrieved": "2025-04-16" }, { "id": "command-a-price-20250315", "url": "https://cohere.com/pricing", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20250315140740/https://cohere.com/pricing", "retrieved": "2025-03-15" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (Mar 13, 2025), context window (256K) and parameter count (111B) — was found in an archived primary source, checked 2026-08-10. Modalities recorded as text-to-text. The primary source states the model is text-only. Capabilities evidenced from the primary sources: agentic: “com/blog\"> < Back to blog Introducing Command A: Max performance, minimal compute Cohere Team Mar 13, 2025 Command A is on par or better than GPT-4o and DeepSeek-V3 ac” | multilingual: “Compared to other leading proprietary and open-weights models Command A delivers maximum performance with minimum hardware costs, excelling on business-critical agentic a” | tool_use: “Tool Use Capabilities: Command A has been specifically trained with conversational tool use capabilities.”." }, "evidence": { "release_date": [ { "value": "2025-03-13", "sources": [ "command-a-s1" ] } ], "context_window": [ { "value": 256000, "sources": [ "command-a-s2" ] } ], "parameter_count": [ { "value": 111000000000, "sources": [ "command-a-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 2.5, "output": 10 }, "currency": "USD", "observed_on": "2025-03-15", "sources": [ "command-a-price-20250315" ], "note": "Base input and output rates as the page read on this date. Fine-tuned and training rates are priced separately and are not recorded here." } ] }, { "id": "gemini-2-5-pro", "model": "Gemini 2.5 Pro", "company": "Google DeepMind", "family": "Gemini", "classification": { "primary_type": "language", "subtype": "reasoning" }, "modalities": { "input": [ "audio", "image", "video", "text" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "function_calling", "structured_output" ], "tags": [ "flagship" ], "note": "The thinking model that led LMArena for months with a one-million-token context.", "events": [ { "type": "announcement", "date": "2025-03-25", "sources": [ "gemini-2-5-pro-s1", "gemini-2-5-pro-s2" ] } ], "specifications": { "language": { "context_window": 1048576, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-2-5-pro-s1", "url": "https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250325170959/https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/", "retrieved": "2025-03-25" }, { "id": "gemini-2-5-pro-s2", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260308075642/https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro", "retrieved": "2026-03-08" }, { "id": "gemini-2-5-pro-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260309160829/https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro?hl=en", "retrieved": "2026-03-09" }, { "id": "gemini-2-5-pro-price-20251103", "url": "https://ai.google.dev/gemini-api/docs/pricing", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20251103074340/https://ai.google.dev/gemini-api/docs/pricing", "retrieved": "2025-11-03" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (Mar 25, 2025) and context window (1 million) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Modalities taken from the documentation's own declaration: \"Inputs Audio, images, video, text, and PDF Output Text\". Capabilities evidenced from the primary sources: reasoning: “Building on this, we recently introduced our first thinking model, Gemini 2.”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2025-03-25", "sources": [ "gemini-2-5-pro-s1" ] } ], "context_window": [ { "value": 1048576, "sources": [ "gemini-2-5-pro-s1", "gemini-2-5-pro-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 1.25, "output": 10 }, "currency": "USD", "observed_on": "2025-11-03", "sources": [ "gemini-2-5-pro-price-20251103" ], "note": "Paid-tier input and output as the page read on this date. The free tier and context-caching rates are priced separately and are not recorded here." } ] }, { "id": "llama-4-scout", "model": "Llama 4 Scout", "company": "Meta AI", "family": "Llama", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision" ], "tags": [], "note": "Meta's smaller Llama 4 mixture-of-experts model, with a 10-million-token context window.", "events": [ { "type": "announcement", "date": "2025-04-05", "sources": [ "llama-4-s1", "llama-4-s2" ] } ], "specifications": { "language": { "context_window": 10000000, "parameter_count": 109000000000 } }, "access": { "open_weights": true, "license": "Llama 4 Community License" }, "sources": [ { "id": "llama-4-s1", "url": "https://ai.meta.com/blog/llama-4-multimodal-intelligence/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250405185132/https://ai.meta.com/blog/llama-4-multimodal-intelligence/", "retrieved": "2025-04-05" }, { "id": "llama-4-s2", "url": "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20250405191842/https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct", "retrieved": "2025-04-05" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Split from the former \"llama-4\" record, which was named for the family while holding one variant's figures. Evidence: \"Llama 4 Scout, is a general purpose model with 17 billion active parameters, 16 experts, and 109 billion total parameters... Llama 4 Scout dramatically increases the supported context length from 128K in Llama 3 to an industry leading 10 million tokens.\" Capabilities evidenced from the primary sources: vision: “As our product workhorse model for general assistant and chat use cases, Llama 4 Maverick is great for precise image understanding and creative writing.”." }, "evidence": { "release_date": [ { "value": "2025-04-05", "sources": [ "llama-4-s1", "llama-4-s2" ] } ], "context_window": [ { "value": 10000000, "sources": [ "llama-4-s1", "llama-4-s2" ] } ], "parameter_count": [ { "value": 109000000000, "sources": [ "llama-4-s1", "llama-4-s2" ] } ] } }, { "id": "llama-4-maverick", "model": "Llama 4 Maverick", "company": "Meta AI", "family": "Llama", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision" ], "tags": [], "note": "Meta's flagship Llama 4 mixture-of-experts model, with 128 experts and 400 billion total parameters.", "events": [ { "type": "announcement", "date": "2025-04-05", "sources": [ "llama-4-s1", "llama-4-s2" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": 400000000000 } }, "access": { "open_weights": true, "license": "Llama 4 Community License" }, "sources": [ { "id": "llama-4-s1", "url": "https://ai.meta.com/blog/llama-4-multimodal-intelligence/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250405185132/https://ai.meta.com/blog/llama-4-multimodal-intelligence/", "retrieved": "2025-04-05" }, { "id": "llama-4-s2", "url": "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20250405191842/https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct", "retrieved": "2025-04-05" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Split from the former \"llama-4\" record, which was named for the family while holding one variant's figures. Evidence: \"Llama 4 Maverick contains 17 billion active parameters, 128 experts, and 400 billion total parameters.\" Capabilities evidenced from the primary sources: vision: “As our product workhorse model for general assistant and chat use cases, Llama 4 Maverick is great for precise image understanding and creative writing.”." }, "evidence": { "release_date": [ { "value": "2025-04-05", "sources": [ "llama-4-s1", "llama-4-s2" ] } ], "context_window": [ { "value": 1000000, "sources": [ "llama-4-s2" ] } ], "parameter_count": [ { "value": 400000000000, "sources": [ "llama-4-s1", "llama-4-s2" ] } ] } }, { "id": "qwen3-235b-a22b", "model": "Qwen3-235B-A22B", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "reasoning" ], "tags": [], "note": "Qwen3's flagship open MoE, with hybrid thinking modes and 22B active parameters.", "events": [ { "type": "announcement", "date": "2025-04-29", "sources": [ "qwen3-s1", "qwen3-s2" ] } ], "specifications": { "language": { "context_window": 131072, "parameter_count": 235000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "qwen3-s1", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250429020635/https://qwenlm.github.io/blog/qwen3/", "retrieved": "2025-04-29" }, { "id": "qwen3-s2", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20250428214435/https://huggingface.co/Qwen/Qwen3-235B-A22B", "retrieved": "2025-04-28" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Split from the former \"qwen3\" record, which was named for the family while holding one variant's figures. Evidence: \"We are open-weighting two MoE models: Qwen3-235B-A22B, a large model with 235 billion total parameters and 22 billion activated parameters, and Qwen3-30B-A3B, a smaller MoE model with 30 billion total parameters and 3 billion activated parameters.\"" }, "evidence": { "release_date": [ { "value": "2025-04-29", "sources": [ "qwen3-s1" ] } ], "context_window": [ { "value": 131072, "sources": [ "qwen3-s2" ] } ], "parameter_count": [ { "value": 235000000000, "sources": [ "qwen3-s1", "qwen3-s2" ] } ] } }, { "id": "qwen3-30b-a3b", "model": "Qwen3-30B-A3B", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "reasoning" ], "tags": [], "note": "The smaller Qwen3 MoE, which the lab said outperformed QwQ-32B with a tenth of the activated parameters.", "events": [ { "type": "announcement", "date": "2025-04-29", "sources": [ "qwen3-s1", "qwen3-s2" ] } ], "specifications": { "language": { "context_window": 131072, "parameter_count": 30000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "qwen3-s1", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250429020635/https://qwenlm.github.io/blog/qwen3/", "retrieved": "2025-04-29" }, { "id": "qwen3-s2", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20250428214435/https://huggingface.co/Qwen/Qwen3-235B-A22B", "retrieved": "2025-04-28" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Split from the former \"qwen3\" record, which was named for the family while holding one variant's figures. Evidence: \"the small MoE model, Qwen3-30B-A3B, outcompetes QwQ-32B with 10 times of activated parameters\"" }, "evidence": { "release_date": [ { "value": "2025-04-29", "sources": [ "qwen3-s1" ] } ], "context_window": [ { "value": 131072, "sources": [ "qwen3-s2" ] } ], "parameter_count": [ { "value": 30000000000, "sources": [ "qwen3-s1" ] } ] } }, { "id": "claude-4-opus", "model": "Claude Opus 4", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "agentic", "vision", "coding" ], "tags": [ "flagship" ], "note": "Anthropic's agentic flagship with computer use and tight Claude Code integration.", "events": [ { "type": "announcement", "date": "2025-05-22", "sources": [ "claude-4-opus-s1", "claude-4-opus-s2" ] } ], "specifications": { "language": { "context_window": 200000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-4-opus-s1", "url": "https://www.anthropic.com/news/claude-4", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250522163841/https://www.anthropic.com/news/claude-4", "retrieved": "2025-05-22" }, { "id": "claude-4-opus-s2", "url": "https://artificialanalysis.ai/models/claude-4-opus", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20250524155453/https://artificialanalysis.ai/models/claude-4-opus", "retrieved": "2025-05-24" }, { "id": "claude-4-opus-s3", "url": "https://docs.anthropic.com/en/docs/about-claude/models/overview", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20250526121527/https://docs.anthropic.com/en/docs/about-claude/models/overview", "retrieved": "2025-05-26" }, { "id": "claude-4-opus-price-20251119", "url": "https://platform.claude.com/docs/en/about-claude/pricing", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20251119190003/https://platform.claude.com/docs/en/about-claude/pricing", "retrieved": "2025-11-19" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Modalities read from Anthropic's model documentation as it stood on 2025-05-26, which lists this model as: “Claude Opus 4 Our most powerful and capable model Text and image input Text output 200k context window Superior reasoning capabilities.” The current documentation no longer lists it, so the contemporaneous snapshot is cited rather than the live page. Capabilities evidenced from the primary sources: coding: “Introducing Claude 4 \\ Anthropic Skip to main content Skip to footer Claude API Solutions Research Commitments Learn News Try Claude Announcements Introducing Claude 4 Ma” | agentic: “Claude Opus 4 is the world’s best coding model, with sustained performance on complex, long-running tasks and agent workflows.” | vision: “Claude Opus 4 Our most powerful and capable model Text and image input Text output 200k context window Superior reasoning capabilities Claude Sonnet 4 High-performance mo”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2025-05-22", "sources": [ "claude-4-opus-s1" ] } ], "context_window": [ { "value": 200000, "sources": [ "claude-4-opus-s3" ] } ] }, "benchmarks": [ { "name": "SWE-bench", "score": 72.5, "evaluation_type": "vendor_reported", "reported_on": "2025-05-22", "sources": [ "claude-4-opus-s1" ] }, { "name": "Terminal-Bench", "score": 43.2, "evaluation_type": "vendor_reported", "reported_on": "2025-05-22", "sources": [ "claude-4-opus-s1" ] } ], "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 15, "output": 75 }, "currency": "USD", "observed_on": "2025-11-19", "sources": [ "claude-4-opus-price-20251119" ], "note": "Base input and output rates as the page read on this date. Cache writes and cache hits are priced separately and are not recorded here." } ] }, { "id": "kimi-k2", "model": "Kimi K2", "company": "Moonshot AI", "family": "Kimi", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "agentic", "coding", "reasoning" ], "tags": [], "note": "A trillion-parameter open MoE with 32B active parameters excelling at code and agents.", "events": [ { "type": "announcement", "date": "2025-07-11", "sources": [ "kimi-k2-s1", "kimi-k2-s2" ] } ], "specifications": { "language": { "context_window": 128000, "parameter_count": 1000000000000 } }, "access": { "open_weights": true, "license": "Modified MIT License" }, "sources": [ { "id": "kimi-k2-s1", "url": "https://moonshotai.github.io/Kimi-K2/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250711154206/https://moonshotai.github.io/Kimi-K2/", "retrieved": "2025-07-11" }, { "id": "kimi-k2-s2", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20250711151833/https://huggingface.co/moonshotai/Kimi-K2-Instruct", "retrieved": "2025-07-11" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (2025-07-11), context window (128K) and parameter count (1T) — was found in an archived primary source, checked 2026-08-10. Modalities recorded as text-to-text. The primary source states the model is text-only. Capabilities evidenced from the primary sources: coding: “Trained with the Muon optimizer, Kimi K2 achieves exceptional performance across frontier knowledge, reasoning, and coding tasks while being meticulously optimized for ag” | agentic: “Trained with the Muon optimizer, Kimi K2 achieves exceptional performance across frontier knowledge, reasoning, and coding tasks while being meticulously optimized for ag” | reasoning: “Evaluation Results Instruction model evaluation results Benchmark Metric Kimi K2 Instruct DeepSeek-V3-0324 Qwen3-235B-A22B (non-thinking) Claude Sonnet 4 (w/o extended th”." }, "evidence": { "release_date": [ { "value": "2025-07-11", "sources": [ "kimi-k2-s2" ] } ], "context_window": [ { "value": 128000, "sources": [ "kimi-k2-s2" ] } ], "parameter_count": [ { "value": 1000000000000, "sources": [ "kimi-k2-s2" ] } ] } }, { "id": "cogvideox-3", "model": "CogVideoX-3", "company": "Zhipu AI", "family": "CogVideoX", "classification": { "primary_type": "video_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "An incremental upgrade to Zhipu's video generation model.", "events": [ { "type": "announcement", "date": "2025-07-15", "sources": [ "cogvideox-3-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "cogvideox-3-s1", "url": "https://docs.z.ai/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260817055142/https://docs.z.ai/release-notes", "retrieved": "2026-08-17" }, { "id": "cogvideox-3-zdocs", "url": "https://docs.z.ai/guides/video/cogvideox-3", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20250730192638/https://docs.z.ai/guides/video/cogvideox-3", "retrieved": "2025-07-30" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2025-07-15", "sources": [ "cogvideox-3-s1" ] } ] } }, { "id": "qwen3-coder-480b-a35b", "model": "Qwen3-Coder-480B-A35B-Instruct", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding", "agentic" ], "tags": [], "note": "A 480B Mixture-of-Experts with 35B active parameters for agentic coding, natively 256K context and extendable to 1M with YaRN.", "events": [ { "type": "announcement", "date": "2025-07-22", "sources": [ "qwen3-coder-480b-a35b-s1" ] } ], "specifications": { "language": { "context_window": 256000, "parameter_count": 480000000000 } }, "access": { "open_weights": true, "license": null }, "sources": [ { "id": "qwen3-coder-480b-a35b-s1", "url": "https://qwenlm.github.io/blog/qwen3-coder/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250722211347/https://qwenlm.github.io/blog/qwen3-coder/", "retrieved": "2025-07-22" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present." }, "evidence": { "release_date": [ { "value": "2025-07-22", "sources": [ "qwen3-coder-480b-a35b-s1" ] } ], "context_window": [ { "value": 256000, "sources": [ "qwen3-coder-480b-a35b-s1" ] } ], "parameter_count": [ { "value": 480000000000, "sources": [ "qwen3-coder-480b-a35b-s1" ] } ] }, "undisclosed": [ "license" ] }, { "id": "glm-4-5", "model": "GLM-4.5", "company": "Zhipu AI", "family": "GLM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "agentic", "coding", "reasoning" ], "tags": [], "note": "A 355B MoE open model that became the first Chinese open model to rival closed frontier models.", "events": [ { "type": "announcement", "date": "2025-07-28", "sources": [ "glm-4-5-s1", "glm-4-5-s2" ] } ], "specifications": { "language": { "context_window": 128000, "parameter_count": 355000000000 } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "glm-4-5-s1", "url": "https://z.ai/blog/glm-4.5", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250728122443/https://z.ai/blog/glm-4.5", "retrieved": "2025-07-28" }, { "id": "glm-4-5-s2", "url": "https://huggingface.co/zai-org/GLM-4.5", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20250728132214/https://huggingface.co/zai-org/GLM-4.5", "retrieved": "2025-07-28" } ], "provenance": { "status": "partially_verified", "confidence": 80, "reason": "Cited to the lab's own announcement and the official model card. Not verified because the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": { "release_date": [ { "value": "2025-07-28", "sources": [ "glm-4-5-s2" ] } ], "parameter_count": [ { "value": 355000000000, "sources": [ "glm-4-5-s2" ] } ] } }, { "id": "qwen-image", "model": "Qwen-Image", "company": "Alibaba Qwen", "family": "Qwen-Image", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": null, "capabilities": [ "image_generation" ], "tags": [], "note": "Alibaba's open image generation model, built around native text rendering inside generated images.", "events": [ { "type": "announcement", "date": "2025-08-04", "sources": [ "qwen-image-s1" ] } ], "specifications": {}, "access": { "open_weights": true, "license": null }, "sources": [ { "id": "qwen-image-s1", "url": "https://qwenlm.github.io/blog/qwen-image/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250804155750/https://qwenlm.github.io/blog/qwen-image/", "retrieved": "2025-08-04" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there. Capabilities evidenced from the primary sources: image_generation: “Strong Cross-Benchmark Performance : Evaluated on multiple public benchmarks, Qwen-Image consistently outperforms existing models across diverse generation and editing tasks, estab”." }, "evidence": { "release_date": [ { "value": "2025-08-04", "sources": [ "qwen-image-s1" ] } ] }, "undisclosed": [ "license" ] }, { "id": "gpt-oss-120b", "model": "gpt-oss-120b", "company": "OpenAI", "family": "GPT-OSS", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "agentic", "coding", "function_calling", "reasoning", "tool_use" ], "tags": [], "note": "OpenAI's larger open-weight model, fitting on a single 80GB GPU.", "events": [ { "type": "announcement", "date": "2025-08-05", "sources": [ "gpt-oss-s1", "gpt-oss-s2", "gpt-oss-s3", "gpt-oss-s4" ] } ], "specifications": { "language": { "context_window": 131072, "parameter_count": 116830000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "gpt-oss-s1", "url": "https://openai.com/index/introducing-gpt-oss/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250805170537/https://openai.com/index/introducing-gpt-oss/", "retrieved": "2025-08-05" }, { "id": "gpt-oss-s2", "url": "https://developers.openai.com/api/docs/models/gpt-oss-120b", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260213072021/https://developers.openai.com/api/docs/models/gpt-oss-120b", "retrieved": "2026-02-13" }, { "id": "gpt-oss-s3", "url": "https://arxiv.org/html/2508.10925v1", "type": "technical_paper", "authority": "primary", "archived_url": "https://web.archive.org/web/20250818033250/https://arxiv.org/html/2508.10925v1", "retrieved": "2025-08-18" }, { "id": "gpt-oss-s4", "url": "https://huggingface.co/openai/gpt-oss-120b", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20250805170129/https://huggingface.co/openai/gpt-oss-120b", "retrieved": "2025-08-05" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Split from the former \"gpt-oss\" record, which was named for the family while holding one variant's figures. Evidence: \"gpt-oss-120b is our most powerful open-weight model, which fits into a single H100 GPU (117B parameters with 5.1B active parameters).\" Capabilities evidenced from the primary sources: tool_use: “Both models also perform strongly on tool use, few-shot function calling, CoT reasoning (as seen in results on the Tau-Bench agentic evaluation suite) and HealthBench (ev” | function_calling: “Both models also perform strongly on tool use, few-shot function calling, CoT reasoning (as seen in results on the Tau-Bench agentic evaluation suite) and HealthBench (ev” | agentic: “Both models also perform strongly on tool use, few-shot function calling, CoT reasoning (as seen in results on the Tau-Bench agentic evaluation suite) and HealthBench (ev” | reasoning: “Similar to the OpenAI o-series reasoning models in the API, the two open-weight models support three reasoning efforts—low, medium, and high—which trade off latency vs.” | coding: “Evaluations We evaluated gpt-oss-120b and gpt-oss-20b across standard academic benchmarks to measure their capabilities in coding, competition math, health, and agentic t”." }, "evidence": { "release_date": [ { "value": "2025-08-05", "sources": [ "gpt-oss-s1", "gpt-oss-s3", "gpt-oss-s4" ] } ], "context_window": [ { "value": 131072, "sources": [ "gpt-oss-s2" ] } ], "parameter_count": [ { "value": 116830000000, "sources": [ "gpt-oss-s1", "gpt-oss-s2", "gpt-oss-s4" ] } ] } }, { "id": "gpt-oss-20b", "model": "gpt-oss-20b", "company": "OpenAI", "family": "GPT-OSS", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "agentic", "coding", "function_calling", "reasoning", "tool_use" ], "tags": [], "note": "OpenAI's smaller open-weight model, running in 16GB of memory.", "events": [ { "type": "announcement", "date": "2025-08-05", "sources": [ "gpt-oss-s1", "gpt-oss-s2", "gpt-oss-s3", "gpt-oss-s4" ] } ], "specifications": { "language": { "context_window": 131072, "parameter_count": 21000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "gpt-oss-s1", "url": "https://openai.com/index/introducing-gpt-oss/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250805170537/https://openai.com/index/introducing-gpt-oss/", "retrieved": "2025-08-05" }, { "id": "gpt-oss-s2", "url": "https://developers.openai.com/api/docs/models/gpt-oss-120b", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260213072021/https://developers.openai.com/api/docs/models/gpt-oss-120b", "retrieved": "2026-02-13" }, { "id": "gpt-oss-s3", "url": "https://arxiv.org/html/2508.10925v1", "type": "technical_paper", "authority": "primary", "archived_url": "https://web.archive.org/web/20250818033250/https://arxiv.org/html/2508.10925v1", "retrieved": "2025-08-18" }, { "id": "gpt-oss-s4", "url": "https://huggingface.co/openai/gpt-oss-120b", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20250805170129/https://huggingface.co/openai/gpt-oss-120b", "retrieved": "2025-08-05" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Split from the former \"gpt-oss\" record, which was named for the family while holding one variant's figures. Evidence: \"gpt-oss-20b ... 21B [total parameters] 3.6B [active]; gpt-oss-20b only requires 16GB.\" Capabilities evidenced from the primary sources: tool_use: “Both models also perform strongly on tool use, few-shot function calling, CoT reasoning (as seen in results on the Tau-Bench agentic evaluation suite) and HealthBench (ev” | function_calling: “Both models also perform strongly on tool use, few-shot function calling, CoT reasoning (as seen in results on the Tau-Bench agentic evaluation suite) and HealthBench (ev” | agentic: “Both models also perform strongly on tool use, few-shot function calling, CoT reasoning (as seen in results on the Tau-Bench agentic evaluation suite) and HealthBench (ev” | reasoning: “Similar to the OpenAI o-series reasoning models in the API, the two open-weight models support three reasoning efforts—low, medium, and high—which trade off latency vs.” | coding: “Evaluations We evaluated gpt-oss-120b and gpt-oss-20b across standard academic benchmarks to measure their capabilities in coding, competition math, health, and agentic t”." }, "evidence": { "release_date": [ { "value": "2025-08-05", "sources": [ "gpt-oss-s1", "gpt-oss-s3", "gpt-oss-s4" ] } ], "context_window": [ { "value": 131072, "sources": [ "gpt-oss-s2" ] } ], "parameter_count": [ { "value": 21000000000, "sources": [ "gpt-oss-s1", "gpt-oss-s4" ] } ] } }, { "id": "gpt-5", "model": "GPT-5", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision", "reasoning", "coding", "structured_output", "function_calling" ], "tags": [ "flagship" ], "note": "OpenAI's unified flagship with integrated reasoning across ChatGPT and the API.", "events": [ { "type": "announcement", "date": "2025-08-07", "sources": [ "gpt-5-s1", "gpt-5-s2" ] } ], "specifications": { "language": { "context_window": 400000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-5-s1", "url": "https://openai.com/index/introducing-gpt-5/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250807174456/https://openai.com/index/introducing-gpt-5/", "retrieved": "2025-08-07" }, { "id": "gpt-5-s2", "url": "https://developers.openai.com/api/docs/models/gpt-5", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260210054829/https://developers.openai.com/api/docs/models/gpt-5", "retrieved": "2026-02-10" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (August 7, 2025) and context window (400,000) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Capabilities evidenced from the primary sources: coding: “GPT‑5 is a significant leap in intelligence over all our previous models, featuring state-of-the-art performance across coding, math, writing, health, visual perception, ” | reasoning: “One unified system GPT‑5 is a unified system with a smart, efficient model that answers most questions, a deeper reasoning model (GPT‑5 thinking) for harder problems, and” | vision: “In our evaluations, GPT‑5 (with thinking) performs better than OpenAI o3 with 50-80% less output tokens across capabilities, including visual reasoning, agentic coding, a”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2025-08-07", "sources": [ "gpt-5-s1", "gpt-5-s2" ] } ], "context_window": [ { "value": 400000, "sources": [ "gpt-5-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 1.25, "output": 10 }, "currency": "USD", "sources": [ "gpt-5-s2" ], "observed_on": "2026-02-10" } ], "benchmarks": [ { "name": "SWE-bench Verified", "score": 74.9, "evaluation_type": "vendor_reported", "reported_on": "2025-08-07", "sources": [ "gpt-5-s1" ] }, { "name": "Aider Polyglot", "score": 88, "evaluation_type": "vendor_reported", "reported_on": "2025-08-07", "sources": [ "gpt-5-s1" ] }, { "name": "MMMU", "score": 84.2, "evaluation_type": "vendor_reported", "reported_on": "2025-08-07", "sources": [ "gpt-5-s1" ] }, { "name": "AIME", "score": 94.6, "evaluation_type": "vendor_reported", "reported_on": "2025-08-07", "sources": [ "gpt-5-s1" ] } ] }, { "id": "glm-4-5v", "model": "GLM-4.5V", "company": "Zhipu AI", "family": "GLM-V", "classification": { "primary_type": "vision", "subtype": null }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "A 100B-scale open vision reasoning model covering a broad range of visual tasks including video.", "events": [ { "type": "announcement", "date": "2025-08-11", "sources": [ "glm-4-5v-s1", "glm-4-5v-s2" ] } ], "specifications": { "language": {} }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "glm-4-5v-s1", "url": "https://docs.z.ai/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260817055142/https://docs.z.ai/release-notes", "retrieved": "2026-08-17" }, { "id": "glm-4-5v-s2", "url": "https://huggingface.co/zai-org/GLM-4.5V", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20250811154730/https://huggingface.co/zai-org/GLM-4.5V", "retrieved": "2025-08-11" }, { "id": "glm-4-5v-zdocs", "url": "https://docs.z.ai/guides/vlm/glm-4.5v", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20251005201008/https://docs.z.ai/guides/vlm/glm-4.5v", "retrieved": "2025-10-05" }, { "id": "glm-4-5v-s4", "url": "https://docs.z.ai/llms-full.txt", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260821052159/https://docs.z.ai/llms-full.txt", "retrieved": "2026-08-21" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to official documentation and the official model card, and every value it asserts was found there. Licence from the model card's own license field on zai-org/GLM-4.5V. Modalities from the model card's own pipeline_tag \"image-text-to-text\" on zai-org/GLM-4.5V." }, "evidence": { "release_date": [ { "value": "2025-08-11", "sources": [ "glm-4-5v-s2" ] } ] } }, { "id": "imagen-4", "model": "Imagen 4", "company": "Google DeepMind", "family": "Imagen", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "Google's text-to-image family, released in Ultra, Standard and Fast tiers.", "events": [ { "type": "announcement", "date": "2025-08-14", "sources": [ "imagen-4-s1", "imagen-4-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "imagen-4-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20250605063552/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2025-06-05" }, { "id": "imagen-4-s2", "url": "https://deepmind.google/models/imagen/", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20250521045650/https://deepmind.google/models/imagen/", "retrieved": "2025-05-21" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be traced: every archived snapshot of those sources predates the release — the newest is 2025-06-05, captured before the page announced it — so the archived copy does not contain the figure. A capture from on or after 2025-08-14 would settle it." } }, { "id": "deepseek-v3-1", "model": "DeepSeek-V3.1", "company": "DeepSeek", "family": "DeepSeek", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "agentic", "function_calling", "reasoning", "tool_use" ], "tags": [], "note": "A hybrid think-and-non-think update doubling context to 128K for the agent era.", "events": [ { "type": "announcement", "date": "2025-08-21", "sources": [ "deepseek-v3-1-s1", "deepseek-v3-1-s2" ] } ], "specifications": { "language": { "context_window": 128000, "parameter_count": 671000000000 } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "deepseek-v3-1-s1", "url": "https://api-docs.deepseek.com/news/news250821", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20250821073920/https://api-docs.deepseek.com/news/news250821", "retrieved": "2025-08-21" }, { "id": "deepseek-v3-1-s2", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.1", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20250821064138/https://huggingface.co/deepseek-ai/DeepSeek-V3.1", "retrieved": "2025-08-21" }, { "id": "deepseek-v3-1-price-20250901", "url": "https://api-docs.deepseek.com/quick_start/pricing/", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20250901020532/https://api-docs.deepseek.com/quick_start/pricing/", "retrieved": "2025-09-01" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (2025-08-21), context window (128K) and parameter count (671B) — was found in an archived primary source, checked 2026-08-10. Modalities recorded as text-to-text. The primary source states the model is text-only. Capabilities evidenced from the primary sources: reasoning: “5 Release 2024/09/05 Context Caching is Available 2024/08/02 New API Features 2024/07/25 API Reference API Guides Reasoning Model (deepseek-reasoner) Multi-round Conversa” | function_calling: “5 Release 2024/09/05 Context Caching is Available 2024/08/02 New API Features 2024/07/25 API Reference API Guides Reasoning Model (deepseek-reasoner) Multi-round Conversa” | tool_use: “DeepSeek-R1-0528 🛠️ Stronger agent skills: Post-training boosts tool use and multi-step agent tasks Try it now — toggle Think/Non-Think via the "DeepThink" but” | agentic: “DeepSeek-R1-0528 🛠️ Stronger agent skills: Post-training boosts tool use and multi-step agent tasks Try it now — toggle Think/Non-Think via the "DeepThink" but”." }, "evidence": { "release_date": [ { "value": "2025-08-21", "sources": [ "deepseek-v3-1-s2" ] } ], "context_window": [ { "value": 128000, "sources": [ "deepseek-v3-1-s1", "deepseek-v3-1-s2" ] } ], "parameter_count": [ { "value": 671000000000, "sources": [ "deepseek-v3-1-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 0.27, "output": 1.1 }, "currency": "USD", "observed_on": "2025-09-01", "sources": [ "deepseek-v3-1-price-20250901" ], "note": "Standard-hours cache-miss input and output as the page read on this date. Cache-hit input and any off-peak discount are priced separately and are not recorded here." } ] }, { "id": "claude-sonnet-4-5", "model": "Claude Sonnet 4.5", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "agentic", "vision", "multilingual" ], "tags": [ "flagship" ], "note": "The best coding model of its time, capable of running agents for 30-plus hours.", "events": [ { "type": "announcement", "date": "2025-09-29", "sources": [ "claude-sonnet-4-5-s1", "claude-sonnet-4-5-s2" ] } ], "specifications": { "language": { "context_window": 200000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-sonnet-4-5-s1", "url": "https://www.anthropic.com/news/claude-sonnet-4-5", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20250929165916/https://www.anthropic.com/news/claude-sonnet-4-5", "retrieved": "2025-09-29" }, { "id": "claude-sonnet-4-5-s2", "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20251122111216/https://platform.claude.com/docs/en/about-claude/models/overview", "retrieved": "2025-11-22" }, { "id": "claude-sonnet-4-5-price-20251119", "url": "https://platform.claude.com/docs/en/about-claude/pricing", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20251119190003/https://platform.claude.com/docs/en/about-claude/pricing", "retrieved": "2025-11-19" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (Sep 29, 2025) and context window (200K) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Capabilities evidenced from the primary sources: multilingual: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.” | vision: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2025-09-29", "sources": [ "claude-sonnet-4-5-s1" ] } ], "context_window": [ { "value": 200000, "sources": [ "claude-sonnet-4-5-s1", "claude-sonnet-4-5-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 3, "output": 15 }, "currency": "USD", "observed_on": "2025-11-19", "sources": [ "claude-sonnet-4-5-price-20251119" ], "note": "Base input and output rates as the page read on this date. Cache writes and cache hits are priced separately and are not recorded here." } ] }, { "id": "glm-4-6", "model": "GLM-4.6", "company": "Zhipu AI", "family": "GLM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding" ], "tags": [], "note": "Zhipu's flagship coding model of the 4.6 generation.", "events": [ { "type": "announcement", "date": "2025-09-30", "sources": [ "glm-4-6-s1", "glm-4-6-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "glm-4-6-s1", "url": "https://docs.z.ai/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260817055142/https://docs.z.ai/release-notes", "retrieved": "2026-08-17" }, { "id": "glm-4-6-s2", "url": "https://huggingface.co/zai-org/GLM-4.6", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20250930155549/https://huggingface.co/zai-org/GLM-4.6", "retrieved": "2025-09-30" }, { "id": "glm-4-6-zdocs", "url": "https://docs.z.ai/guides/llm/glm-4.6", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260112194543/https://docs.z.ai/guides/llm/glm-4.6", "retrieved": "2026-01-12" }, { "id": "glm-4-6-s4", "url": "https://docs.z.ai/llms-full.txt", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260821052159/https://docs.z.ai/llms-full.txt", "retrieved": "2026-08-21" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to official documentation and the official model card, and every value it asserts was found there. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present. Licence from the model card's own license field on zai-org/GLM-4.6." }, "evidence": { "release_date": [ { "value": "2025-09-30", "sources": [ "glm-4-6-s2" ] } ] } }, { "id": "sora-2", "model": "Sora 2", "company": "OpenAI", "family": "Sora", "classification": { "primary_type": "video_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "OpenAI's video generation model, the second generation of Sora.", "events": [ { "type": "announcement", "date": "2025-09-30", "sources": [ "sora-2-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "sora-2-s1", "url": "https://openai.com/index/sora-2/", "type": "official_announcement", "authority": "primary", "archived_url": "https://archive.md/20260410121703/https://openai.com/index/sora-2/", "retrieved": "2026-04-10" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." } }, { "id": "claude-haiku-4-5", "model": "Claude Haiku 4.5", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "reasoning" ], "tags": [], "note": "Anthropic's small model tier. The API id claude-haiku-4-5-20251001 carries a snapshot date two weeks before the announcement; the release date is the one Anthropic states.", "events": [ { "type": "announcement", "date": "2025-10-15", "sources": [ "claude-haiku-4-5-s1" ] } ], "specifications": { "language": { "context_window": 200000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-haiku-4-5-s1", "url": "https://www.anthropic.com/news/claude-haiku-4-5", "type": "official_announcement", "authority": "primary", "archived_url": "https://archive.md/20251015195111/https://www.anthropic.com/news/claude-haiku-4-5", "retrieved": "2025-10-15" }, { "id": "claude-haiku-4-5-adocs", "url": "https://platform.claude.com/docs/en/about-claude/models/overview.md", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260817055519/https://platform.claude.com/docs/en/about-claude/models/overview.md", "retrieved": "2026-08-17" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement and official documentation. Not verified because the release date and the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 1, "output": 5 }, "currency": "USD", "observed_on": "2026-08-17", "sources": [ "claude-haiku-4-5-adocs" ], "note": "Standard tier. Anthropic prices batch and cached tokens differently." } ] }, { "id": "gpt-5-1", "model": "GPT-5.1", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "reasoning" ], "tags": [ "flagship" ], "note": "A five-model refresh of the GPT-5 line, shipped in two waves in November 2025.", "events": [ { "type": "announcement", "date": "2025-11-12", "sources": [ "gpt-5-1-s1", "gpt-5-1-s2" ] } ], "specifications": { "language": { "context_window": 400000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-5-1-s1", "url": "https://en.wikipedia.org/wiki/GPT-5.1", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20251113191350/https://en.wikipedia.org/wiki/GPT-5.1", "retrieved": "2025-11-13" }, { "id": "gpt-5-1-s2", "url": "https://developers.openai.com/api/docs/models/gpt-5.1", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260213212943/https://developers.openai.com/api/docs/models/gpt-5.1", "retrieved": "2026-02-13" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities taken from the documentation's own declaration: \"Input Text, image Output Text\". Capabilities evidenced from the primary sources: reasoning: “Reasoning token support”." }, "undisclosed": [ "parameter_count" ], "evidence": { "context_window": [ { "value": 400000, "sources": [ "gpt-5-1-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 1.25, "output": 10 }, "currency": "USD", "sources": [ "gpt-5-1-s2" ], "observed_on": "2026-02-13" } ] }, { "id": "gemini-3-pro", "model": "Gemini 3 Pro", "company": "Google DeepMind", "family": "Gemini", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "capabilities": [ "vision", "coding", "agentic", "reasoning", "function_calling", "structured_output" ], "tags": [ "flagship" ], "note": "Google's natively multimodal reasoning flagship with DeepThink and 1M context.", "events": [ { "type": "announcement", "date": "2025-11-18", "sources": [ "gemini-3-pro-s1", "gemini-3-pro-s2" ] } ], "specifications": { "language": { "context_window": 1048576, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-3-pro-s1", "url": "https://blog.google/products/gemini/gemini-3/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20251118160715/https://blog.google/products/gemini/gemini-3/", "retrieved": "2025-11-18" }, { "id": "gemini-3-pro-s2", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-Pro-Model-Card.pdf", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20251118111103/https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-Pro-Model-Card.pdf", "retrieved": "2025-11-18" }, { "id": "gemini-3-pro-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3-pro-preview?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260219050535/https://ai.google.dev/gemini-api/docs/models/gemini-3-pro-preview?hl=en", "retrieved": "2026-02-19" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (Nov 18, 2025) and context window (1 million) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Modalities read from the launch announcement, which states: “built from the start to seamlessly synthesize information about any topic across multiple modalities, including text, images, video, audio and code.” Output is recorded as text: no image, audio or video generation is claimed anywhere in the primary sources, which labs advertise prominently when present. Capabilities evidenced from the primary sources: coding: “Gemini 3 Pro outperforms previous models in reasoning, multimodality, and coding benchmarks.” | agentic: “Gemini 2 laid the foundation for agentic capabilities and pushed the frontiers on reasoning and thinking , helping with more complex tasks and ideas, leading to Gemini 2.”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2025-11-18", "sources": [ "gemini-3-pro-s1" ] } ], "context_window": [ { "value": 1048576, "sources": [ "gemini-3-pro-s1", "gemini-3-pro-s2" ] } ] }, "benchmarks": [ { "name": "SWE-bench Verified", "score": 76.2, "evaluation_type": "vendor_reported", "reported_on": "2025-11-18", "sources": [ "gemini-3-pro-s1" ] }, { "name": "Terminal-Bench", "score": 54.2, "evaluation_type": "vendor_reported", "reported_on": "2025-11-18", "sources": [ "gemini-3-pro-s1" ] }, { "name": "MMMU", "score": 81, "evaluation_type": "vendor_reported", "reported_on": "2025-11-18", "sources": [ "gemini-3-pro-s1" ] }, { "name": "GPQA Diamond", "score": 91.9, "evaluation_type": "vendor_reported", "reported_on": "2025-11-18", "sources": [ "gemini-3-pro-s1" ] }, { "name": "Humanity’s Last Exam", "score": 37.5, "evaluation_type": "vendor_reported", "reported_on": "2025-11-18", "sources": [ "gemini-3-pro-s1" ] } ] }, { "id": "olmo-3-7b", "model": "Olmo 3 7B", "company": "Allen Institute for AI", "family": "Olmo", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding", "function_calling" ], "tags": [], "note": "Part of Ai2's fully-open Olmo 3 family, released with every checkpoint, data point and dependency used to build it.", "events": [ { "type": "announcement", "date": "2025-11-20", "sources": [ "olmo-3-7b-s1", "olmo-3-7b-s2", "olmo-3-7b-s3" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 7000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "olmo-3-7b-s1", "url": "https://allenai.org/blog/olmo3", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20251120152225/https://allenai.org/blog/olmo3", "retrieved": "2025-11-20" }, { "id": "olmo-3-7b-s2", "url": "https://arxiv.org/abs/2512.13961", "type": "technical_paper", "authority": "primary", "archived_url": "https://web.archive.org/web/20251217032856/https://arxiv.org/abs/2512.13961", "retrieved": "2025-12-17" }, { "id": "olmo-3-7b-s3", "url": "https://huggingface.co/allenai/Olmo-3-7B-Instruct", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20251120144811/https://huggingface.co/allenai/Olmo-3-7B-Instruct", "retrieved": "2025-11-20" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — release date and parameter count — was found in an archived primary source. Context window is not recorded: none of the cited sources states one, so nothing is claimed about it. Capabilities evidenced from the primary sources: coding: “On math benchmarks, Olmo 3-Think (7B) matches Qwen 3 8B on MATH and comes within a few points on AIME 2024 and 2025, and also leads all comparison models on HumanEvalPlus” | function_calling: “Olmo 3 model construction targets long-context reasoning, function calling, coding, instruction following, general chat, and knowledge recall.”." }, "evidence": { "release_date": [ { "value": "2025-11-20", "sources": [ "olmo-3-7b-s1", "olmo-3-7b-s3" ] } ], "parameter_count": [ { "value": 7000000000, "sources": [ "olmo-3-7b-s1", "olmo-3-7b-s2", "olmo-3-7b-s3" ] } ] }, "undisclosed": [ "context_window" ] }, { "id": "olmo-3-32b", "model": "Olmo 3 32B", "company": "Allen Institute for AI", "family": "Olmo", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding", "function_calling" ], "tags": [ "flagship" ], "note": "Part of Ai2's fully-open Olmo 3 family, released with every checkpoint, data point and dependency used to build it.", "events": [ { "type": "announcement", "date": "2025-11-20", "sources": [ "olmo-3-32b-s1", "olmo-3-32b-s2", "olmo-3-32b-s3" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 32000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "olmo-3-32b-s1", "url": "https://allenai.org/blog/olmo3", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20251120152225/https://allenai.org/blog/olmo3", "retrieved": "2025-11-20" }, { "id": "olmo-3-32b-s2", "url": "https://arxiv.org/abs/2512.13961", "type": "technical_paper", "authority": "primary", "archived_url": "https://web.archive.org/web/20251217032856/https://arxiv.org/abs/2512.13961", "retrieved": "2025-12-17" }, { "id": "olmo-3-32b-s3", "url": "https://huggingface.co/allenai/Olmo-3-32B-Think", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20251120151210/https://huggingface.co/allenai/Olmo-3-32B-Think", "retrieved": "2025-11-20" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — release date and parameter count — was found in an archived primary source. Context window is not recorded: none of the cited sources states one, so nothing is claimed about it. Capabilities evidenced from the primary sources: coding: “On math benchmarks, Olmo 3-Think (7B) matches Qwen 3 8B on MATH and comes within a few points on AIME 2024 and 2025, and also leads all comparison models on HumanEvalPlus” | function_calling: “Olmo 3 model construction targets long-context reasoning, function calling, coding, instruction following, general chat, and knowledge recall.”." }, "evidence": { "release_date": [ { "value": "2025-11-20", "sources": [ "olmo-3-32b-s1", "olmo-3-32b-s3" ] } ], "parameter_count": [ { "value": 32000000000, "sources": [ "olmo-3-32b-s1", "olmo-3-32b-s2", "olmo-3-32b-s3" ] } ] }, "undisclosed": [ "context_window" ] }, { "id": "claude-opus-4-5", "model": "Claude Opus 4.5", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "agentic", "vision", "multilingual" ], "tags": [ "flagship" ], "note": "Anthropic's most capable model for coding, agents, and computer use.", "events": [ { "type": "announcement", "date": "2025-11-24", "sources": [ "claude-opus-4-5-s1", "claude-opus-4-5-s2" ] } ], "specifications": { "language": { "context_window": 200000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-opus-4-5-s1", "url": "https://www.anthropic.com/news/claude-opus-4-5", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20251124190104/https://www.anthropic.com/news/claude-opus-4-5", "retrieved": "2025-11-24" }, { "id": "claude-opus-4-5-s2", "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20251123190158/https://platform.claude.com/docs/en/about-claude/models/overview", "retrieved": "2025-11-23" }, { "id": "claude-opus-4-5-price-20251201", "url": "https://platform.claude.com/docs/en/about-claude/pricing", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20251201041432/https://platform.claude.com/docs/en/about-claude/pricing", "retrieved": "2025-12-01" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (Nov 24, 2025) and context window (200K) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Capabilities evidenced from the primary sources: multilingual: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.” | vision: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2025-11-24", "sources": [ "claude-opus-4-5-s1" ] } ], "context_window": [ { "value": 200000, "sources": [ "claude-opus-4-5-s1", "claude-opus-4-5-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 5, "output": 25 }, "currency": "USD", "observed_on": "2025-12-01", "sources": [ "claude-opus-4-5-price-20251201" ], "note": "Base input and output rates as the page read on this date. Cache writes and cache hits are priced separately and are not recorded here." } ] }, { "id": "flux-2-dev", "model": "FLUX.2 [dev]", "company": "Black Forest Labs", "family": "FLUX", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "A 32B open-weight image generation and editing model, combining text-to-image with editing from multiple reference images in one checkpoint.", "events": [ { "type": "announcement", "date": "2025-11-25", "sources": [ "flux-2-dev-s1" ] } ], "specifications": {}, "access": { "open_weights": true, "license": null }, "sources": [ { "id": "flux-2-dev-s1", "url": "https://bfl.ai/blog/flux-2", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20251125162530/https://bfl.ai/blog/flux-2", "retrieved": "2025-11-25" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2025-11-25", "sources": [ "flux-2-dev-s1" ] } ] } }, { "id": "flux-2-pro", "model": "FLUX.2 [pro]", "company": "Black Forest Labs", "family": "FLUX", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "The managed endpoint variant, which the lab positions against closed frontier image models.", "events": [ { "type": "announcement", "date": "2025-11-25", "sources": [ "flux-2-pro-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "flux-2-pro-s1", "url": "https://bfl.ai/blog/flux-2", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20251125162530/https://bfl.ai/blog/flux-2", "retrieved": "2025-11-25" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2025-11-25", "sources": [ "flux-2-pro-s1" ] } ] } }, { "id": "mistral-large-3", "model": "Mistral Large 3", "company": "Mistral AI", "family": "Mistral", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision", "multilingual" ], "tags": [], "note": "Mistral's most capable model, a sparse mixture-of-experts under Apache 2.0.", "events": [ { "type": "announcement", "date": "2025-12-02", "sources": [ "mistral-3-s1", "mistral-3-s2" ] } ], "specifications": { "language": { "context_window": 256000, "parameter_count": 675000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "mistral-3-s1", "url": "https://techcrunch.com/2025/12/02/mistral-closes-in-on-big-ai-rivals-with-mistral-3-open-weight-frontier-and-small-models/", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20251202155302/https://techcrunch.com/2025/12/02/mistral-closes-in-on-big-ai-rivals-with-mistral-3-open-weight-frontier-and-small-models/", "retrieved": "2025-12-02" }, { "id": "mistral-3-s2", "url": "https://mistral.ai/news/mistral-3/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20251202152420/https://mistral.ai/news/mistral-3", "retrieved": "2025-12-02" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Split from the former \"mistral-3\" record, which was named for the family while holding one variant's figures. Evidence: \"Mistral Large 3 – our most capable model to date – a sparse mixture-of-experts trained with 41B active and 675B total parameters.\" Capabilities evidenced from the primary sources: multilingual: “Introducing Mistral 3 | Mistral AI Products Solutions Research Resources Pricing Company Try AI Studio Talk to sales Mistral 3 Mistral 3 Mistral 3 Mistral 3 Mistral 3 Mis”." }, "evidence": { "parameter_count": [ { "value": 675000000000, "sources": [ "mistral-3-s2" ] } ] } }, { "id": "ministral-3-3b", "model": "Ministral 3 3B", "company": "Mistral AI", "family": "Mistral", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision", "multilingual" ], "tags": [], "note": "The smallest Ministral 3 model, for cost-sensitive and edge deployment.", "events": [ { "type": "announcement", "date": "2025-12-02", "sources": [ "mistral-3-s1", "mistral-3-s2" ] } ], "specifications": { "language": { "context_window": 256000, "parameter_count": 3000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "mistral-3-s1", "url": "https://techcrunch.com/2025/12/02/mistral-closes-in-on-big-ai-rivals-with-mistral-3-open-weight-frontier-and-small-models/", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20251202155302/https://techcrunch.com/2025/12/02/mistral-closes-in-on-big-ai-rivals-with-mistral-3-open-weight-frontier-and-small-models/", "retrieved": "2025-12-02" }, { "id": "mistral-3-s2", "url": "https://mistral.ai/news/mistral-3/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20251202152420/https://mistral.ai/news/mistral-3", "retrieved": "2025-12-02" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Split from the former \"mistral-3\" record, which was named for the family while holding one variant's figures. Evidence: \"Ministral 3 series, available in three model sizes: 3B, 8B, and 14B parameters.\" Capabilities evidenced from the primary sources: multilingual: “Introducing Mistral 3 | Mistral AI Products Solutions Research Resources Pricing Company Try AI Studio Talk to sales Mistral 3 Mistral 3 Mistral 3 Mistral 3 Mistral 3 Mis”." }, "evidence": { "parameter_count": [ { "value": 3000000000, "sources": [ "mistral-3-s2" ] } ] } }, { "id": "ministral-3-8b", "model": "Ministral 3 8B", "company": "Mistral AI", "family": "Mistral", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision", "multilingual" ], "tags": [], "note": "The mid-size Ministral 3 model.", "events": [ { "type": "announcement", "date": "2025-12-02", "sources": [ "mistral-3-s1", "mistral-3-s2" ] } ], "specifications": { "language": { "context_window": 256000, "parameter_count": 8000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "mistral-3-s1", "url": "https://techcrunch.com/2025/12/02/mistral-closes-in-on-big-ai-rivals-with-mistral-3-open-weight-frontier-and-small-models/", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20251202155302/https://techcrunch.com/2025/12/02/mistral-closes-in-on-big-ai-rivals-with-mistral-3-open-weight-frontier-and-small-models/", "retrieved": "2025-12-02" }, { "id": "mistral-3-s2", "url": "https://mistral.ai/news/mistral-3/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20251202152420/https://mistral.ai/news/mistral-3", "retrieved": "2025-12-02" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Split from the former \"mistral-3\" record, which was named for the family while holding one variant's figures. Evidence: \"Ministral 3 series, available in three model sizes: 3B, 8B, and 14B parameters.\" Capabilities evidenced from the primary sources: multilingual: “Introducing Mistral 3 | Mistral AI Products Solutions Research Resources Pricing Company Try AI Studio Talk to sales Mistral 3 Mistral 3 Mistral 3 Mistral 3 Mistral 3 Mis”." }, "evidence": { "parameter_count": [ { "value": 8000000000, "sources": [ "mistral-3-s2" ] } ] } }, { "id": "ministral-3-14b", "model": "Ministral 3 14B", "company": "Mistral AI", "family": "Mistral", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision", "multilingual" ], "tags": [], "note": "The largest Ministral 3 model.", "events": [ { "type": "announcement", "date": "2025-12-02", "sources": [ "mistral-3-s1", "mistral-3-s2" ] } ], "specifications": { "language": { "context_window": 256000, "parameter_count": 14000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "mistral-3-s1", "url": "https://techcrunch.com/2025/12/02/mistral-closes-in-on-big-ai-rivals-with-mistral-3-open-weight-frontier-and-small-models/", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20251202155302/https://techcrunch.com/2025/12/02/mistral-closes-in-on-big-ai-rivals-with-mistral-3-open-weight-frontier-and-small-models/", "retrieved": "2025-12-02" }, { "id": "mistral-3-s2", "url": "https://mistral.ai/news/mistral-3/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20251202152420/https://mistral.ai/news/mistral-3", "retrieved": "2025-12-02" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Split from the former \"mistral-3\" record, which was named for the family while holding one variant's figures. Evidence: \"Ministral 3 series, available in three model sizes: 3B, 8B, and 14B parameters.\" Capabilities evidenced from the primary sources: multilingual: “Introducing Mistral 3 | Mistral AI Products Solutions Research Resources Pricing Company Try AI Studio Talk to sales Mistral 3 Mistral 3 Mistral 3 Mistral 3 Mistral 3 Mis”." }, "evidence": { "parameter_count": [ { "value": 14000000000, "sources": [ "mistral-3-s2" ] } ] } }, { "id": "glm-4-6v", "model": "GLM-4.6V", "company": "Zhipu AI", "family": "GLM-V", "classification": { "primary_type": "vision", "subtype": null }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "The 4.6-generation multimodal model, with improved vision understanding.", "events": [ { "type": "announcement", "date": "2025-12-08", "sources": [ "glm-4-6v-s1", "glm-4-6v-s2" ] } ], "specifications": { "language": {} }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "glm-4-6v-s1", "url": "https://docs.z.ai/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260817055142/https://docs.z.ai/release-notes", "retrieved": "2026-08-17" }, { "id": "glm-4-6v-s2", "url": "https://huggingface.co/zai-org/GLM-4.6V", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20251208114911/https://huggingface.co/zai-org/GLM-4.6V", "retrieved": "2025-12-08" }, { "id": "glm-4-6v-zdocs", "url": "https://docs.z.ai/guides/vlm/glm-4.6v", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20251208114929/https://docs.z.ai/guides/vlm/glm-4.6v", "retrieved": "2025-12-08" }, { "id": "glm-4-6v-s4", "url": "https://docs.z.ai/llms-full.txt", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260821052159/https://docs.z.ai/llms-full.txt", "retrieved": "2026-08-21" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to official documentation and the official model card, and every value it asserts was found there. Licence from the model card's own license field on zai-org/GLM-4.6V. Modalities from the model card's own pipeline_tag \"image-text-to-text\" on zai-org/GLM-4.6V." }, "evidence": { "release_date": [ { "value": "2025-12-08", "sources": [ "glm-4-6v-s2" ] } ] } }, { "id": "glm-asr-2512", "model": "GLM-ASR-2512", "company": "Zhipu AI", "family": "GLM-ASR", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "A speech recognition model the lab reports at a 0.0717 character error rate.", "events": [ { "type": "announcement", "date": "2025-12-10", "sources": [ "glm-asr-2512-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "glm-asr-2512-s1", "url": "https://docs.z.ai/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260817055142/https://docs.z.ai/release-notes", "retrieved": "2026-08-17" }, { "id": "glm-asr-2512-zdocs", "url": "https://docs.z.ai/guides/audio/glm-asr-2512", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20251213061254/https://docs.z.ai/guides/audio/glm-asr-2512", "retrieved": "2025-12-13" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2025-12-10", "sources": [ "glm-asr-2512-s1" ] } ] } }, { "id": "gpt-5-2", "model": "GPT-5.2", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "reasoning" ], "tags": [ "flagship" ], "note": "Instant, Thinking and Pro modes, plus a GPT-5.2-Codex variant for coding.", "events": [ { "type": "announcement", "date": "2025-12-11", "sources": [ "gpt-5-2-s1", "gpt-5-2-s2" ] } ], "specifications": { "language": { "context_window": 400000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-5-2-s1", "url": "https://en.wikipedia.org/wiki/GPT-5.2", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20251218083248/https://en.wikipedia.org/wiki/GPT-5.2", "retrieved": "2025-12-18" }, { "id": "gpt-5-2-s2", "url": "https://developers.openai.com/api/docs/models/gpt-5.2", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260210005456/https://developers.openai.com/api/docs/models/gpt-5.2", "retrieved": "2026-02-10" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (2025-12-11) and context window (400,000) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Modalities taken from the documentation's own declaration: \"Input Text, image Output Text\". Capabilities evidenced from the primary sources: reasoning: “Reasoning token support”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2025-12-11", "sources": [ "gpt-5-2-s2" ] } ], "context_window": [ { "value": 400000, "sources": [ "gpt-5-2-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 1.75, "output": 14 }, "currency": "USD", "sources": [ "gpt-5-2-s2" ], "observed_on": "2026-02-10" } ] }, { "id": "mimo-v2-flash", "model": "MiMo-V2-Flash", "company": "Xiaomi", "family": "MiMo", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "coding", "agentic" ], "tags": [], "note": "Xiaomi released MiMo-V2-Flash with a hybrid architecture using a 1:5 ratio of global attention to sliding-window attention (window size 128), native 32K context and training extended to 256K (262,144 in the published config). Xiaomi's changelog records two later updates, on 2026-01-12 and 2026-02-04, strengthening general and coding performance in Thinking mode.", "events": [ { "type": "announcement", "date": "2025-12-16", "sources": [ "mimo-v2-flash-s1", "mimo-v2-flash-s2" ] }, { "type": "major_update", "date": "2026-01-12", "sources": [ "mimo-v2-flash-s1" ] }, { "type": "major_update", "date": "2026-02-04", "sources": [ "mimo-v2-flash-s1" ] } ], "specifications": { "language": { "context_window": 262144, "parameter_count": null } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "mimo-v2-flash-s1", "url": "https://mimo.mi.com/docs/en-US/updates/model", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260616040133/https://mimo.mi.com/docs/en-US/updates/model", "retrieved": "2026-06-16" }, { "id": "mimo-v2-flash-s2", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20251217022202/https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash", "retrieved": "2025-12-17" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`. Capabilities evidenced from the primary sources: coding: “mimo-v2-flash-0204 mimo-v2-flash-0112 mimo-v2-flash SWE-Bench Verified Non-Thinking 73.” | agentic: “Post-Training Technical Highlights MiMo-V2-Flash leverages a post-training pipeline designed to maximize reasoning and agentic capabilities through innovative distillation and rein”." }, "evidence": { "release_date": [ { "value": "2025-12-16", "sources": [ "mimo-v2-flash-s1", "mimo-v2-flash-s2" ] } ], "context_window": [ { "value": 262144, "sources": [ "mimo-v2-flash-s1", "mimo-v2-flash-s2" ] } ] } }, { "id": "glm-4-7", "model": "GLM-4.7", "company": "Zhipu AI", "family": "GLM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "coding", "reasoning" ], "tags": [], "note": "The 4.7 foundation model, with improvements to coding, reasoning and agentic behaviour.", "events": [ { "type": "announcement", "date": "2025-12-22", "sources": [ "glm-4-7-s1", "glm-4-7-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "glm-4-7-s1", "url": "https://docs.z.ai/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260817055142/https://docs.z.ai/release-notes", "retrieved": "2026-08-17" }, { "id": "glm-4-7-s2", "url": "https://huggingface.co/zai-org/GLM-4.7", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20251222173319/https://huggingface.co/zai-org/GLM-4.7", "retrieved": "2025-12-22" }, { "id": "glm-4-7-zdocs", "url": "https://docs.z.ai/guides/llm/glm-4.7", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20251222173159/https://docs.z.ai/guides/llm/glm-4.7", "retrieved": "2025-12-22" }, { "id": "glm-4-7-s4", "url": "https://docs.z.ai/llms-full.txt", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260821052159/https://docs.z.ai/llms-full.txt", "retrieved": "2026-08-21" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to official documentation and the official model card, and every value it asserts was found there. Licence from the model card's own license field on zai-org/GLM-4.7." }, "evidence": { "release_date": [ { "value": "2025-12-22", "sources": [ "glm-4-7-s2" ] } ] } }, { "id": "nemotron-3-nano-30b-a3b", "model": "Nemotron 3 Nano 30B-A3B", "company": "NVIDIA", "family": "Nemotron", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "A 31.6B open Mixture-of-Experts hybrid Mamba-Transformer aimed at agentic reasoning on a single GPU.", "events": [ { "type": "announcement", "date": "2025-12-23", "sources": [ "nemotron-3-nano-30b-a3b-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 31600000000 } }, "access": { "open_weights": true, "license": null }, "sources": [ { "id": "nemotron-3-nano-30b-a3b-s1", "url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Nano-Technical-Report.pdf", "type": "technical_paper", "authority": "primary", "archived_url": "https://web.archive.org/web/20251217015106/https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Nano-Technical-Report.pdf", "retrieved": "2025-12-17" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the technical paper. Not verified because the release date could not be traced: every archived snapshot of those sources predates the release — the newest is 2025-12-17, captured before the page announced it — so the archived copy does not contain the figure. A capture from on or after 2025-12-23 would settle it. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present." }, "evidence": { "parameter_count": [ { "value": 31600000000, "sources": [ "nemotron-3-nano-30b-a3b-s1" ] } ] }, "undisclosed": [ "context_window", "license" ] }, { "id": "glm-image", "model": "GLM-Image", "company": "Zhipu AI", "family": "GLM-Image", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": { "input": [ "text" ], "output": [ "image" ] }, "capabilities": [], "tags": [], "note": "An image generation model built on a multimodal architecture and trained on domestic compute.", "events": [ { "type": "announcement", "date": "2026-01-14", "sources": [ "glm-image-s1", "glm-image-s2" ] } ], "specifications": {}, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "glm-image-s1", "url": "https://docs.z.ai/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260817055142/https://docs.z.ai/release-notes", "retrieved": "2026-08-17" }, { "id": "glm-image-s2", "url": "https://huggingface.co/zai-org/GLM-Image", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260114043506/https://huggingface.co/zai-org/GLM-Image", "retrieved": "2026-01-14" }, { "id": "glm-image-zdocs", "url": "https://docs.z.ai/guides/image/glm-image", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260330123428/https://docs.z.ai/guides/image/glm-image", "retrieved": "2026-03-30" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation and the official model card. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Licence from the model card's own license field on zai-org/GLM-Image. Modalities from the model card's own pipeline_tag \"text-to-image\" on zai-org/GLM-Image." }, "evidence": {} }, { "id": "glm-4-7-flash", "model": "GLM-4.7-Flash", "company": "Zhipu AI", "family": "GLM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "The lightweight free-tier counterpart to GLM-4.7.", "events": [ { "type": "announcement", "date": "2026-01-19", "sources": [ "glm-4-7-flash-s1", "glm-4-7-flash-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "glm-4-7-flash-s1", "url": "https://docs.z.ai/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260817055142/https://docs.z.ai/release-notes", "retrieved": "2026-08-17" }, { "id": "glm-4-7-flash-s2", "url": "https://huggingface.co/zai-org/GLM-4.7-Flash", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260119150010/https://huggingface.co/zai-org/GLM-4.7-Flash", "retrieved": "2026-01-19" }, { "id": "glm-4-7-flash-s3", "url": "https://docs.z.ai/llms-full.txt", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260821052159/https://docs.z.ai/llms-full.txt", "retrieved": "2026-08-21" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to official documentation and the official model card, and every value it asserts was found there. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present. Licence from the model card's own license field on zai-org/GLM-4.7-Flash." }, "evidence": { "release_date": [ { "value": "2026-01-19", "sources": [ "glm-4-7-flash-s2" ] } ] } }, { "id": "minimax-speech-2-8", "model": "MiniMax Speech 2.8", "company": "MiniMax", "family": "MiniMax Speech", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "MiniMax's speech model line.", "events": [ { "type": "announcement", "date": "2026-01-23", "sources": [ "minimax-speech-2-8-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "minimax-speech-2-8-s1", "url": "https://www.minimax.io/news/minimax-speech-28", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260419231337/https://www.minimax.io/news/minimax-speech-28", "retrieved": "2026-04-19" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": {} }, { "id": "qwen3-max-thinking", "model": "Qwen3-Max-Thinking", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "reasoning" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "reasoning" ], "tags": [], "note": "The reasoning configuration of Qwen3-Max, served through the API as qwen3-max-2026-01-23.", "events": [ { "type": "announcement", "date": "2026-01-23", "sources": [ "qwen3-max-thinking-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "qwen3-max-thinking-s1", "url": "https://qwen.ai/blog?id=qwen3-max-thinking", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260126154449/https://qwen.ai/blog?id=qwen3-max-thinking", "retrieved": "2026-01-26" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present." }, "evidence": {}, "undisclosed": [ "context_window", "parameter_count" ] }, { "id": "grok-imagine-video", "model": "Grok Imagine Video", "company": "xAI", "family": "Grok Imagine", "classification": { "primary_type": "video_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "xAI launched the Grok Imagine API around what it called its most powerful video-audio generative model: video generation from text or an image, and video editing.", "events": [ { "type": "announcement", "date": "2026-01-28", "sources": [ "grok-imagine-video-s1", "grok-imagine-video-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "grok-imagine-video-s1", "url": "https://x.ai/news/grok-imagine-api", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260129105331/https://x.ai/news/grok-imagine-api", "retrieved": "2026-01-29" }, { "id": "grok-imagine-video-s2", "url": "https://docs.x.ai/developers/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260208020350/https://docs.x.ai/developers/release-notes", "retrieved": "2026-02-08" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-01-28", "sources": [ "grok-imagine-video-s1" ] } ] } }, { "id": "qwen3-coder-next", "model": "Qwen3-Coder-Next", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding" ], "tags": [], "note": "An open coding model in the Qwen3 line, released between the Qwen3.5 and Qwen3.6 generations.", "events": [ { "type": "announcement", "date": "2026-02-02", "sources": [ "qwen3-coder-next-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": true, "license": null }, "sources": [ { "id": "qwen3-coder-next-s1", "url": "https://qwen.ai/blog?id=qwen3-coder-next", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260203161636/https://qwen.ai/blog?id=qwen3-coder-next", "retrieved": "2026-02-03" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present." }, "evidence": {}, "undisclosed": [ "context_window", "license", "parameter_count" ] }, { "id": "glm-ocr", "model": "GLM-OCR", "company": "Zhipu AI", "family": "GLM-OCR", "classification": { "primary_type": "vision", "subtype": null }, "modalities": { "input": [ "image" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "A compact optical character recognition model built on Zhipu's own CogViT.", "events": [ { "type": "announcement", "date": "2026-02-03", "sources": [ "glm-ocr-s1", "glm-ocr-s2" ] } ], "specifications": {}, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "glm-ocr-s1", "url": "https://docs.z.ai/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260817055142/https://docs.z.ai/release-notes", "retrieved": "2026-08-17" }, { "id": "glm-ocr-s2", "url": "https://huggingface.co/zai-org/GLM-OCR", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260203014757/https://huggingface.co/zai-org/GLM-OCR", "retrieved": "2026-02-03" }, { "id": "glm-ocr-zdocs", "url": "https://docs.z.ai/guides/vlm/glm-ocr", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260203013651/https://docs.z.ai/guides/vlm/glm-ocr", "retrieved": "2026-02-03" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation and the official model card. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Licence from the model card's own license field on zai-org/GLM-OCR. Modalities from the model card's own pipeline_tag \"image-to-text\" on zai-org/GLM-OCR." }, "evidence": {} }, { "id": "claude-opus-4-6", "model": "Claude Opus 4.6", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "agentic", "vision", "multilingual" ], "tags": [ "flagship" ], "note": "The first Opus release of 2026 and the baseline Opus 4.7 was measured against.", "events": [ { "type": "announcement", "date": "2026-02-05", "sources": [ "claude-opus-4-6-s1", "claude-opus-4-6-s2" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-opus-4-6-s1", "url": "https://www.anthropic.com/news/claude-opus-4-6", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260205175758/https://www.anthropic.com/news/claude-opus-4-6", "retrieved": "2026-02-05" }, { "id": "claude-opus-4-6-s2", "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260204091134/https://platform.claude.com/docs/en/about-claude/models/overview", "retrieved": "2026-02-04" }, { "id": "claude-opus-4-6-price-20260301", "url": "https://platform.claude.com/docs/en/about-claude/pricing", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260301150515/https://platform.claude.com/docs/en/about-claude/pricing", "retrieved": "2026-03-01" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (Feb 5, 2026) and context window (1M) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Capabilities evidenced from the primary sources: multilingual: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.” | vision: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2026-02-05", "sources": [ "claude-opus-4-6-s1" ] } ], "context_window": [ { "value": 1000000, "sources": [ "claude-opus-4-6-s1", "claude-opus-4-6-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 5, "output": 25 }, "currency": "USD", "observed_on": "2026-03-01", "sources": [ "claude-opus-4-6-price-20260301" ], "note": "Base input and output rates as the page read on this date. Cache writes and cache hits are priced separately and are not recorded here." } ] }, { "id": "gpt-5-3-codex", "model": "GPT-5.3-Codex", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "agentic", "vision" ], "tags": [], "note": "A coding-specialised model able to search repositories, run terminal commands and debug, positioned against Claude Opus 4.6.", "events": [ { "type": "announcement", "date": "2026-02-05", "sources": [ "gpt-5-3-codex-s1", "gpt-5-3-codex-s2" ] } ], "specifications": { "language": { "context_window": 400000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-5-3-codex-s1", "url": "https://en.wikipedia.org/wiki/GPT-5.3-Codex", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20260501204242/https://en.wikipedia.org/wiki/GPT-5.3-Codex", "retrieved": "2026-05-01" }, { "id": "gpt-5-3-codex-s2", "url": "https://developers.openai.com/api/docs/models/gpt-5.3-codex", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260226193922/https://developers.openai.com/api/docs/models/gpt-5.3-codex", "retrieved": "2026-02-26" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date and the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities read from OpenAI's own model documentation, which states: “Modalities Text Input and output Image Input only Audio Not supported Video Not supported.” The snapshot cited was moved from 2026-02-16 to 2026-02-26: the earlier capture caught only the page shell before its content rendered, so it evidenced nothing it was cited for. Capabilities evidenced from the primary sources: vision: “Modalities Text Input and output Image Input only Audio Not supported Video Not supported”." }, "undisclosed": [ "parameter_count" ], "evidence": { "context_window": [ { "value": 400000, "sources": [ "gpt-5-3-codex-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 1.75, "output": 14 }, "currency": "USD", "observed_on": "2026-02-26", "sources": [ "gpt-5-3-codex-s2" ] } ] }, { "id": "glm-5", "model": "GLM-5", "company": "Zhipu AI", "family": "GLM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding", "agentic" ], "tags": [], "note": "Built for complex system engineering and long-range agent tasks, which the lab frames as a shift from coding to engineering.", "events": [ { "type": "announcement", "date": "2026-02-12", "sources": [ "glm-5-s1", "glm-5-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "glm-5-s1", "url": "https://docs.z.ai/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260817055142/https://docs.z.ai/release-notes", "retrieved": "2026-08-17" }, { "id": "glm-5-s2", "url": "https://huggingface.co/zai-org/GLM-5", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260212032142/https://huggingface.co/zai-org/GLM-5", "retrieved": "2026-02-12" }, { "id": "glm-5-zdocs", "url": "https://docs.z.ai/guides/llm/glm-5", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260212034837/https://docs.z.ai/guides/llm/glm-5", "retrieved": "2026-02-12" }, { "id": "glm-5-s4", "url": "https://docs.z.ai/llms-full.txt", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260821052159/https://docs.z.ai/llms-full.txt", "retrieved": "2026-08-21" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation and the official model card. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. The primary source states the model is text-only. Licence from the model card's own license field on zai-org/GLM-5." }, "evidence": {} }, { "id": "seedance-2-0", "model": "Seedance 2.0", "company": "ByteDance", "family": "Seedance", "classification": { "primary_type": "video_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "The 2.0 generation of ByteDance's video generation model.", "events": [ { "type": "announcement", "date": "2026-02-12", "sources": [ "seedance-2-0-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "seedance-2-0-s1", "url": "https://seed.bytedance.com/blog/seedance-2-0-official-launch", "type": "official_announcement", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." } }, { "id": "seedream-5-0-lite", "model": "Seedream 5.0 Lite", "company": "ByteDance", "family": "Seedream", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "The lighter Seedream 5.0 variant, positioned around more deliberate generation.", "events": [ { "type": "announcement", "date": "2026-02-13", "sources": [ "seedream-5-0-lite-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "seedream-5-0-lite-s1", "url": "https://seed.bytedance.com/blog/deeper-thinking-more-accurate-generation-introducing-seedream-5-0-lite", "type": "official_announcement", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." } }, { "id": "qwen-3-5", "model": "Qwen3.5-397B-A17B", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [ "flagship" ], "note": "The first Qwen3.5 model released open-weight: a native vision-language MoE with 397B total and 17B active parameters.", "events": [ { "type": "announcement", "date": "2026-02-14", "sources": [ "qwen-3-5-s0", "qwen-3-5-s2" ] } ], "specifications": { "language": { "context_window": 262144, "parameter_count": 397000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "qwen-3-5-s0", "url": "https://qwen.ai/blog?id=qwen3.5", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260216094454/https://qwen.ai/blog?id=qwen3.5", "retrieved": "2026-02-16" }, { "id": "qwen-3-5-s2", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260216104038/https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "retrieved": "2026-02-16" } ], "provenance": { "status": "partially_verified", "confidence": 80, "reason": "Cited to the lab's own announcement and the official model card. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Licence from the model card's own license field on Qwen/Qwen3.5-397B-A17B." }, "evidence": { "context_window": [ { "value": 262144, "sources": [ "qwen-3-5-s2" ] } ], "parameter_count": [ { "value": 397000000000, "sources": [ "qwen-3-5-s2" ] } ] } }, { "id": "seed-2-0", "model": "Seed2.0", "company": "ByteDance", "family": "Seed", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [], "tags": [], "note": "The 2.0 foundation model, with upgraded multimodal understanding and stronger agent performance.", "events": [ { "type": "announcement", "date": "2026-02-14", "sources": [ "seed-2-0-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "seed-2-0-s1", "url": "https://seed.bytedance.com/blog/seed-2-0-official-launch", "type": "official_announcement", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." } }, { "id": "claude-sonnet-4-6", "model": "Claude Sonnet 4.6", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "agentic", "vision", "multilingual" ], "tags": [], "note": "The Sonnet-tier counterpart to Opus 4.6.", "events": [ { "type": "announcement", "date": "2026-02-17", "sources": [ "claude-sonnet-4-6-s1", "claude-sonnet-4-6-s2" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-sonnet-4-6-s1", "url": "https://www.anthropic.com/news/claude-sonnet-4-6", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260217175906/https://www.anthropic.com/news/claude-sonnet-4-6", "retrieved": "2026-02-17" }, { "id": "claude-sonnet-4-6-s2", "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260217170302/https://platform.claude.com/docs/en/about-claude/models/overview", "retrieved": "2026-02-17" }, { "id": "claude-sonnet-4-6-price-20260301", "url": "https://platform.claude.com/docs/en/about-claude/pricing", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260301150515/https://platform.claude.com/docs/en/about-claude/pricing", "retrieved": "2026-03-01" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (Feb 17, 2026) and context window (1M) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Capabilities evidenced from the primary sources: multilingual: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.” | vision: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2026-02-17", "sources": [ "claude-sonnet-4-6-s1" ] } ], "context_window": [ { "value": 1000000, "sources": [ "claude-sonnet-4-6-s1", "claude-sonnet-4-6-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 3, "output": 15 }, "currency": "USD", "observed_on": "2026-03-01", "sources": [ "claude-sonnet-4-6-price-20260301" ], "note": "Base input and output rates as the page read on this date. Cache writes and cache hits are priced separately and are not recorded here." } ] }, { "id": "nano-banana-2", "model": "Nano Banana 2", "company": "Google DeepMind", "family": "Nano Banana", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "image_generation" ], "tags": [], "note": "The image generation and editing model Google ships inside Gemini, and the most widely used of its image line.", "events": [ { "type": "announcement", "date": "2026-02-26", "sources": [ "nano-banana-2-s1", "nano-banana-2-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "nano-banana-2-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260227144730/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2026-02-27" }, { "id": "nano-banana-2-s2", "url": "https://deepmind.google/models/nano-banana/", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null }, { "id": "nano-banana-2-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-image?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260530143309/https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-image?hl=en", "retrieved": "2026-05-30" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to official documentation, and every value it asserts was found there. Modalities recorded as text-to-text. The primary source states the model is text-only." }, "evidence": { "release_date": [ { "value": "2026-02-26", "sources": [ "nano-banana-2-s1" ] } ] } }, { "id": "grok-4-20", "model": "Grok 4.20", "company": "xAI", "family": "Grok", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "reasoning" ], "tags": [], "note": "xAI's release notes list Grok 4.20 as live in March 2026 and give no day; the date is recorded at month precision rather than inferred. It is served as grok-4.20-0309-reasoning and grok-4.20-0309-non-reasoning — the 0309 is a snapshot identifier, not a stated release date.", "events": [ { "type": "announcement", "date": "2026-03", "sources": [ "grok-4-20-s1", "grok-4-20-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "grok-4-20-s1", "url": "https://docs.x.ai/developers/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260217134702/https://docs.x.ai/developers/release-notes", "retrieved": "2026-02-17" }, { "id": "grok-4-20-s2", "url": "https://docs.x.ai/developers/models", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260304091958/https://docs.x.ai/developers/models", "retrieved": "2026-03-04" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": {}, "undisclosed": [ "context_window", "parameter_count" ] }, { "id": "grok-4-20-multi-agent", "model": "Grok 4.20 Multi-agent", "company": "xAI", "family": "Grok", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "agentic", "reasoning" ], "tags": [], "note": "Released alongside Grok 4.20 in March 2026, per xAI's release notes, which give no day. Served as grok-4.20-multi-agent-0309.", "events": [ { "type": "announcement", "date": "2026-03", "sources": [ "grok-4-20-multi-agent-s1", "grok-4-20-multi-agent-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "grok-4-20-multi-agent-s1", "url": "https://docs.x.ai/developers/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260217134702/https://docs.x.ai/developers/release-notes", "retrieved": "2026-02-17" }, { "id": "grok-4-20-multi-agent-s2", "url": "https://docs.x.ai/developers/models", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260304091958/https://docs.x.ai/developers/models", "retrieved": "2026-03-04" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": {}, "undisclosed": [ "context_window", "parameter_count" ] }, { "id": "gpt-5-4", "model": "GPT-5.4", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "vision" ], "tags": [ "flagship" ], "note": "OpenAI's frontier model with 1M context and state-of-the-art coding and computer use.", "events": [ { "type": "announcement", "date": "2026-03-05", "sources": [ "gpt-5-4-s1", "gpt-5-4-s2" ] } ], "specifications": { "language": { "context_window": 1050000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-5-4-s1", "url": "https://techcrunch.com/2026/03/05/openai-launches-gpt-5-4-with-pro-and-thinking-versions/", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20260305182156/https://techcrunch.com/2026/03/05/openai-launches-gpt-5-4-with-pro-and-thinking-versions/", "retrieved": "2026-03-05" }, { "id": "gpt-5-4-s2", "url": "https://developers.openai.com/api/docs/models/gpt-5.4", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260305185140/https://developers.openai.com/api/docs/models/gpt-5.4", "retrieved": "2026-03-05" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (2026-03-05) and context window (1,050,000) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Modalities taken from the documentation's own declaration: \"Input Text, image Output Text\". Capabilities evidenced from the primary sources: vision: “Modalities Text Input and output Image Input only Audio Not supported Video Not supported” | reasoning: “Reasoning token support”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2026-03-05", "sources": [ "gpt-5-4-s2" ] } ], "context_window": [ { "value": 1050000, "sources": [ "gpt-5-4-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 2.5, "output": 15 }, "currency": "USD", "sources": [ "gpt-5-4-s2" ], "observed_on": "2026-03-05" } ] }, { "id": "nemotron-3-super", "model": "Nemotron 3 Super", "company": "NVIDIA", "family": "Nemotron", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "coding", "reasoning" ], "tags": [], "note": "A 120B open hybrid Mamba-Transformer MoE for agentic reasoning, with a one-million-token context.", "events": [ { "type": "announcement", "date": "2026-03-11", "sources": [ "nemotron-3-super-s1" ] }, { "type": "weights_availability", "date": "2026-03-11", "sources": [ "nemotron-3-super-s1" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": true, "license": null }, "sources": [ { "id": "nemotron-3-super-s1", "url": "https://developer.nvidia.com/blog/introducing-nemotron-3-super-an-open-hybrid-mamba-transformer-moe-for-agentic-reasoning/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260311192531/https://developer.nvidia.com/blog/introducing-nemotron-3-super-an-open-hybrid-mamba-transformer-moe-for-agentic-reasoning/", "retrieved": "2026-03-11" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there. Capabilities evidenced from the primary sources: coding: “For instance, in software development, simple merge requests can be addressed by Nemotron 3 Nano while complex coding tasks that require deeper understanding of the code base can b” | reasoning: “Open training and evaluation infrastructure NVIDIA publishes development techniques and tools, giving researchers and enterprises the flexibility to customize Nemotron 3 Super or b”." }, "evidence": { "release_date": [ { "value": "2026-03-11", "sources": [ "nemotron-3-super-s1" ] } ], "context_window": [ { "value": 1000000, "sources": [ "nemotron-3-super-s1" ] } ] }, "undisclosed": [ "license", "parameter_count" ] }, { "id": "glm-5-turbo", "model": "GLM-5-Turbo", "company": "Zhipu AI", "family": "GLM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "agentic" ], "tags": [], "note": "A high-throughput variant aimed at stability and efficiency in long-chain agent workloads.", "events": [ { "type": "announcement", "date": "2026-03-15", "sources": [ "glm-5-turbo-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "glm-5-turbo-s1", "url": "https://docs.z.ai/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260817055142/https://docs.z.ai/release-notes", "retrieved": "2026-08-17" }, { "id": "glm-5-turbo-zdocs", "url": "https://docs.z.ai/guides/llm/glm-5-turbo", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260316015247/https://docs.z.ai/guides/llm/glm-5-turbo", "retrieved": "2026-03-16" }, { "id": "glm-5-turbo-s3", "url": "https://docs.z.ai/llms-full.txt", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260821052159/https://docs.z.ai/llms-full.txt", "retrieved": "2026-08-21" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." } }, { "id": "mistral-small-4", "model": "Mistral Small 4", "company": "Mistral AI", "family": "Mistral", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision", "reasoning", "agentic", "coding" ], "tags": [], "note": "A 119B open mixture-of-experts model unifying chat, coding, and agentic tasks.", "events": [ { "type": "announcement", "date": "2026-03-16", "sources": [ "mistral-small-4-s1" ] } ], "specifications": { "language": { "context_window": 256000, "parameter_count": 119000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "mistral-small-4-s1", "url": "https://mistral.ai/news/mistral-small-4/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260316210415/https://mistral.ai/news/mistral-small-4", "retrieved": "2026-03-16" } ], "provenance": { "status": "partially_verified", "confidence": 80, "reason": "Cited to the lab's own announcement. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": { "context_window": [ { "value": 256000, "sources": [ "mistral-small-4-s1" ] } ], "parameter_count": [ { "value": 119000000000, "sources": [ "mistral-small-4-s1" ] } ] } }, { "id": "mimo-v2-pro", "model": "MiMo-V2-Pro", "company": "Xiaomi", "family": "MiMo", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [], "tags": [], "note": "Xiaomi released MiMo-V2-Pro with a hybrid architecture using a 1:7 ratio of global attention to sliding-window attention, 1T total parameters and 42B active. No weights repository was published alongside it.", "events": [ { "type": "announcement", "date": "2026-03-18", "sources": [ "mimo-v2-pro-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 1000000000000 } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mimo-v2-pro-s1", "url": "https://mimo.mi.com/docs/en-US/updates/model", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260616040133/https://mimo.mi.com/docs/en-US/updates/model", "retrieved": "2026-06-16" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-03-18", "sources": [ "mimo-v2-pro-s1" ] } ], "parameter_count": [ { "value": 1000000000000, "sources": [ "mimo-v2-pro-s1" ] } ] } }, { "id": "lyria-3", "model": "Lyria 3", "company": "Google DeepMind", "family": "Lyria", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "Google's music generation model, shipped as a 30-second clip preview and a full-length song preview.", "events": [ { "type": "announcement", "date": "2026-03-25", "sources": [ "lyria-3-s1", "lyria-3-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "lyria-3-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20250605063552/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2025-06-05" }, { "id": "lyria-3-s2", "url": "https://deepmind.google/models/lyria/", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20250521051108/https://deepmind.google/models/lyria/", "retrieved": "2025-05-21" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be traced: every archived snapshot of those sources predates the release — the newest is 2025-06-05, captured before the page announced it — so the archived copy does not contain the figure. A capture from on or after 2026-03-25 would settle it." } }, { "id": "gemini-3-1-flash-live", "model": "Gemini 3.1 Flash Live", "company": "Google DeepMind", "family": "Gemini Live", "classification": { "primary_type": "audio", "subtype": null }, "modalities": { "input": [ "text", "audio", "video" ], "output": [ "text", "audio" ] }, "capabilities": [ "reasoning", "function_calling" ], "tags": [], "note": "An audio-to-audio model for real-time dialogue.", "events": [ { "type": "announcement", "date": "2026-03-26", "sources": [ "gemini-3-1-flash-live-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-3-1-flash-live-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20250605063552/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2025-06-05" }, { "id": "gemini-3-1-flash-live-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-live-preview?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260327085354/https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-live-preview?hl=en", "retrieved": "2026-03-27" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." } }, { "id": "veo-3", "model": "Veo 3", "company": "Google DeepMind", "family": "Veo", "classification": { "primary_type": "video_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "Google's video generation model, which produces cinematic video with synchronised audio.", "events": [ { "type": "announcement", "date": "2026-03-31", "sources": [ "veo-3-s1", "veo-3-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "veo-3-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260404053513/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2026-04-04" }, { "id": "veo-3-s2", "url": "https://deepmind.google/models/veo/", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20250909114804/https://deepmind.google/models/veo/", "retrieved": "2025-09-09" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." } }, { "id": "glm-5v-turbo", "model": "GLM-5V-Turbo", "company": "Zhipu AI", "family": "GLM-V", "classification": { "primary_type": "vision", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "Native multimodal understanding for vision-based coding and agent workflows.", "events": [ { "type": "announcement", "date": "2026-04-01", "sources": [ "glm-5v-turbo-s1" ] } ], "specifications": { "language": {} }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "glm-5v-turbo-s1", "url": "https://docs.z.ai/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260817055142/https://docs.z.ai/release-notes", "retrieved": "2026-08-17" }, { "id": "glm-5v-turbo-zdocs", "url": "https://docs.z.ai/guides/vlm/glm-5v-turbo", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260402035459/https://docs.z.ai/guides/vlm/glm-5v-turbo", "retrieved": "2026-04-02" }, { "id": "glm-5v-turbo-s3", "url": "https://docs.z.ai/llms-full.txt", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260821052159/https://docs.z.ai/llms-full.txt", "retrieved": "2026-08-21" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." } }, { "id": "qwen-3-6-plus", "model": "Qwen3.6-Plus", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "agentic" ], "tags": [], "note": "An API-only agentic upgrade over Qwen3.5, with a one-million-token context.", "events": [ { "type": "announcement", "date": "2026-04-02", "sources": [ "qwen-3-6-plus-s1" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "qwen-3-6-plus-s1", "url": "https://qwen.ai/blog?id=qwen3.6", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260402050028/https://qwen.ai/blog?id=qwen3.6", "retrieved": "2026-04-02" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement. Not verified because the release date and the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present." }, "evidence": {}, "undisclosed": [ "parameter_count" ] }, { "id": "glm-5-1", "model": "GLM-5.1", "company": "Zhipu AI", "family": "GLM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "agentic" ], "tags": [], "note": "Built for long-horizon work, which the lab says can run independently for up to eight hours in a single session.", "events": [ { "type": "announcement", "date": "2026-04-07", "sources": [ "glm-5-1-s1", "glm-5-1-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "glm-5-1-s1", "url": "https://docs.z.ai/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260817055142/https://docs.z.ai/release-notes", "retrieved": "2026-08-17" }, { "id": "glm-5-1-s2", "url": "https://huggingface.co/zai-org/GLM-5.1", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260407161846/https://huggingface.co/zai-org/GLM-5.1", "retrieved": "2026-04-07" }, { "id": "glm-5-1-zdocs", "url": "https://docs.z.ai/guides/llm/glm-5.1", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260407175312/https://docs.z.ai/guides/llm/glm-5.1", "retrieved": "2026-04-07" }, { "id": "glm-5-1-s4", "url": "https://docs.z.ai/llms-full.txt", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260821052159/https://docs.z.ai/llms-full.txt", "retrieved": "2026-08-21" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to official documentation and the official model card, and every value it asserts was found there. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present. Licence from the model card's own license field on zai-org/GLM-5.1." }, "evidence": { "release_date": [ { "value": "2026-04-07", "sources": [ "glm-5-1-s2" ] } ] } }, { "id": "llama-5", "model": "Llama 5", "company": "Meta AI", "family": "Llama", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [], "tags": [ "flagship" ], "note": "Meta's first MoE flagship, marking a major architecture shift for the open family.", "events": [ { "type": "announcement", "date": "2026-04-08", "sources": [ "llama-5-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": true, "license": null }, "sources": [ { "id": "llama-5-s1", "url": "https://en.wikipedia.org/wiki/Llama_(language_model)", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20260415051351/https://en.wikipedia.org/wiki/Llama_(language_model)", "retrieved": "2026-04-15" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "No primary source. The release is recorded from news reporting only, so it cannot be verified until the lab's own announcement is cited." } }, { "id": "muse-spark", "model": "Muse Spark", "company": "Meta AI", "family": "Muse", "classification": { "primary_type": "language", "subtype": "reasoning" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision", "reasoning", "tool_use", "agentic" ], "tags": [ "flagship" ], "note": "Meta's first closed frontier model, from Meta Superintelligence Labs, shipped alongside the open-weights Llama 5.", "events": [ { "type": "announcement", "date": "2026-04-08", "sources": [ "muse-spark-s1", "muse-spark-s2", "muse-spark-s3" ] } ], "specifications": { "language": { "context_window": 262144, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "muse-spark-s1", "url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260408163043/https://ai.meta.com/blog/introducing-muse-spark-msl/", "retrieved": "2026-04-08" }, { "id": "muse-spark-s2", "url": "https://en.wikipedia.org/wiki/Muse_Spark", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20260718013514/https://en.wikipedia.org/wiki/Muse_Spark", "retrieved": "2026-07-18" }, { "id": "muse-spark-s3", "url": "https://artificialanalysis.ai/models/muse-spark", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20260408164722/https://artificialanalysis.ai/models/muse-spark", "retrieved": "2026-04-08" } ], "provenance": { "status": "partially_verified", "confidence": 80, "reason": "Cited to the lab's own announcement. Not verified because the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2026-04-08", "sources": [ "muse-spark-s1" ] } ] }, "benchmarks": [ { "name": "Humanity’s Last Exam", "score": 58, "evaluation_type": "vendor_reported", "reported_on": "2026-04-08", "sources": [ "muse-spark-s1" ] } ] }, { "id": "mai-image-2-efficient", "model": "MAI-Image-2-Efficient", "company": "Microsoft", "family": "MAI-Image", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "A cost-reduced tier of the MAI-Image-2 line, positioned at flagship quality for lower cost.", "events": [ { "type": "announcement", "date": "2026-04-14", "sources": [ "mai-image-2-efficient-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mai-image-2-efficient-s1", "url": "https://microsoft.ai/news/mai-image-2-efficient/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260420081724/https://microsoft.ai/news/mai-image-2-efficient/", "retrieved": "2026-04-20" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2026-04-14", "sources": [ "mai-image-2-efficient-s1" ] } ] } }, { "id": "qwen-3-6-35b-a3b", "model": "Qwen3.6-35B-A3B", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "agentic", "coding", "reasoning" ], "tags": [], "note": "A 35-billion-parameter mixture-of-experts Qwen release with 3B active parameters.", "events": [ { "type": "announcement", "date": "2026-04-15", "sources": [ "qwen-3-6-35b-a3b-s1", "qwen-3-6-35b-a3b-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 35000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "qwen-3-6-35b-a3b-s1", "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260416133036/https://huggingface.co/Qwen/Qwen3.6-35B-A3B", "retrieved": "2026-04-16" }, { "id": "qwen-3-6-35b-a3b-s2", "url": "https://qwen.ai/blog?id=qwen3.6", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260413115807/https://qwen.ai/blog?id=qwen3.6", "retrieved": "2026-04-13" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the official model card and the lab's own announcement, and every value it asserts was found there. Modalities recorded as text-to-text. The primary source states the model is text-only. Capabilities evidenced from the primary sources: agentic: “6 via the Chat Completions API Agentic Usage Qwen-Agent Qwen Code Processing Ultra-Long Texts Best Practices Citation Qwen3.” | coding: “6-35B-A3B}: Agentic Coding Power, Now Open to All}, url = {https://qwen.”." }, "evidence": { "release_date": [ { "value": "2026-04-15", "sources": [ "qwen-3-6-35b-a3b-s1" ] } ], "parameter_count": [ { "value": 35000000000, "sources": [ "qwen-3-6-35b-a3b-s1" ] } ] } }, { "id": "claude-opus-4-7", "model": "Claude Opus 4.7", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "agentic", "vision", "multilingual" ], "tags": [ "flagship" ], "note": "Anthropic's most capable general model with 87.6% on SWE-bench Verified.", "events": [ { "type": "announcement", "date": "2026-04-16", "sources": [ "claude-opus-4-7-s1", "claude-opus-4-7-s2" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-opus-4-7-s1", "url": "https://www.anthropic.com/news/claude-opus-4-7", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260416142724/https://www.anthropic.com/news/claude-opus-4-7", "retrieved": "2026-04-16" }, { "id": "claude-opus-4-7-s2", "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260415205431/https://platform.claude.com/docs/en/about-claude/models/overview", "retrieved": "2026-04-15" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (Apr 16, 2026) and context window (1M) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Capabilities evidenced from the primary sources: agentic: “Across our agentic reasoning over data benchmarks, it is the best-performing Claude model for enterprise document analysis.” | multilingual: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.” | vision: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2026-04-16", "sources": [ "claude-opus-4-7-s1" ] } ], "context_window": [ { "value": 1000000, "sources": [ "claude-opus-4-7-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 5, "output": 25 }, "currency": "USD", "sources": [ "claude-opus-4-7-s1" ], "observed_on": "2026-04-16" } ] }, { "id": "grok-stt", "model": "Grok STT", "company": "xAI", "family": "Grok Voice", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "xAI's speech-to-text model (grok-stt), announced with the Speech to Text API: transcription in 25 languages, batch and streaming.", "events": [ { "type": "announcement", "date": "2026-04-17", "sources": [ "grok-stt-s1", "grok-stt-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "grok-stt-s1", "url": "https://x.ai/news/grok-stt-and-tts-apis", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260418002920/https://x.ai/news/grok-stt-and-tts-apis", "retrieved": "2026-04-18" }, { "id": "grok-stt-s2", "url": "https://docs.x.ai/developers/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260413113022/https://docs.x.ai/developers/release-notes", "retrieved": "2026-04-13" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-04-17", "sources": [ "grok-stt-s1" ] } ] } }, { "id": "grok-tts", "model": "Grok TTS", "company": "xAI", "family": "Grok Voice", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "xAI's text-to-speech model (grok-tts). xAI's release notes list the Text-to-Speech API as available in March 2026; the dated announcement of both speech APIs followed on 2026-04-17.", "events": [ { "type": "announcement", "date": "2026-04-17", "sources": [ "grok-tts-s1", "grok-tts-s2" ] }, { "type": "api_availability", "date": "2026-03", "sources": [ "grok-tts-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "grok-tts-s1", "url": "https://x.ai/news/grok-stt-and-tts-apis", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260418002920/https://x.ai/news/grok-stt-and-tts-apis", "retrieved": "2026-04-18" }, { "id": "grok-tts-s2", "url": "https://docs.x.ai/developers/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260413113022/https://docs.x.ai/developers/release-notes", "retrieved": "2026-04-13" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-04-17", "sources": [ "grok-tts-s1" ] } ] } }, { "id": "qwen-3-6-27b", "model": "Qwen3.6-27B", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "agentic", "coding", "reasoning" ], "tags": [], "note": "The 27-billion-parameter dense model in the Qwen3.6 release.", "events": [ { "type": "announcement", "date": "2026-04-21", "sources": [ "qwen-3-6-27b-s1", "qwen-3-6-27b-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 27000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "qwen-3-6-27b-s1", "url": "https://huggingface.co/Qwen/Qwen3.6-27B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260422132942/https://huggingface.co/Qwen/Qwen3.6-27B", "retrieved": "2026-04-22" }, { "id": "qwen-3-6-27b-s2", "url": "https://qwen.ai/blog?id=qwen3.6", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260422072805/https://qwen.ai/blog%3Fid%3Dqwen3.6", "retrieved": "2026-04-22" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the official model card and the lab's own announcement, and every value it asserts was found there. Modalities recorded as text-to-text. The primary source states the model is text-only. Capabilities evidenced from the primary sources: agentic: “6 via the Chat Completions API Agentic Usage Qwen-Agent Qwen Code Processing Ultra-Long Texts Best Practices Citation Qwen3.” | coding: “6-27B}: Flagship-Level Coding in a {27B} Dense Model}, author = {{Qwen Team}}, month = {April}, year = {2026}, url = {https://qwen.”." }, "evidence": { "release_date": [ { "value": "2026-04-21", "sources": [ "qwen-3-6-27b-s1" ] } ], "parameter_count": [ { "value": 27000000000, "sources": [ "qwen-3-6-27b-s1" ] } ] } }, { "id": "gpt-5-5", "model": "GPT-5.5", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "agentic" ], "tags": [ "flagship" ], "note": "Codenamed Spud; the mid-cycle step between GPT-5.4 and the GPT-5.6 family.", "events": [ { "type": "announcement", "date": "2026-04-23", "sources": [ "gpt-5-5-s1", "gpt-5-5-s2", "gpt-5-5-s3" ] } ], "specifications": { "language": { "context_window": 1050000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-5-5-s1", "url": "https://openai.com/index/introducing-gpt-5-5/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260423180657/https://openai.com/index/introducing-gpt-5-5/", "retrieved": "2026-04-23" }, { "id": "gpt-5-5-s2", "url": "https://en.wikipedia.org/wiki/GPT-5.5", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20260425005034/https://en.wikipedia.org/wiki/GPT-5.5", "retrieved": "2026-04-25" }, { "id": "gpt-5-5-s3", "url": "https://developers.openai.com/api/docs/models/gpt-5.5", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260424185633/https://developers.openai.com/api/docs/models/gpt-5.5", "retrieved": "2026-04-24" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (April 23, 2026) and context window (1,050,000) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Modalities taken from the documentation's own declaration: \"Input Text, image Output Text\". Capabilities evidenced from the primary sources: agentic: “1% - Model capabilities OpenAI is building the global infrastructure for agentic AI, making it possible for people and businesses around the world to get work done with A” | reasoning: “1% Evals of GPT were run with reasoning effort set to xhigh and were conducted in a research environment, which may provide slightly different output from production Chat”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2026-04-23", "sources": [ "gpt-5-5-s1", "gpt-5-5-s3" ] } ], "context_window": [ { "value": 1050000, "sources": [ "gpt-5-5-s3" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 5, "output": 30 }, "currency": "USD", "sources": [ "gpt-5-5-s3" ], "observed_on": "2026-04-24" } ], "benchmarks": [ { "name": "OSWorld", "score": 78.7, "evaluation_type": "vendor_reported", "reported_on": "2026-04-23", "sources": [ "gpt-5-5-s1" ] } ] }, { "id": "seed3d-2-0", "model": "Seed3D 2.0", "company": "ByteDance", "family": "Seed3D", "classification": { "primary_type": "3d", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "ByteDance's 3D generation model, released with higher precision and broader usability.", "events": [ { "type": "announcement", "date": "2026-04-23", "sources": [ "seed3d-2-0-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "seed3d-2-0-s1", "url": "https://seed.bytedance.com/blog/seed3d-2-0-released-higher-precision-and-greater-usability", "type": "official_announcement", "authority": "primary", "archived_url": "https://archive.md/20260423051829/https://seed.bytedance.com/blog/seed3d-2-0-released-higher-precision-and-greater-usability", "retrieved": "2026-04-23" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." } }, { "id": "mimo-v2-5", "model": "MiMo-V2.5", "company": "Xiaomi", "family": "MiMo", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "agentic" ], "tags": [], "note": "Xiaomi released MiMo-V2.5 with native full-modal perception of images, video, audio and text, a 1M-token context (1,048,576 in the published config) and native agent execution.", "events": [ { "type": "announcement", "date": "2026-04-23", "sources": [ "mimo-v2-5-s1", "mimo-v2-5-s2" ] } ], "specifications": { "language": { "context_window": 1048576, "parameter_count": null } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "mimo-v2-5-s1", "url": "https://mimo.mi.com/docs/en-US/updates/model", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260616040133/https://mimo.mi.com/docs/en-US/updates/model", "retrieved": "2026-06-16" }, { "id": "mimo-v2-5-s2", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260427173640/https://huggingface.co/XiaomiMiMo/MiMo-V2.5", "retrieved": "2026-04-27" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-04-23", "sources": [ "mimo-v2-5-s1" ] } ], "context_window": [ { "value": 1048576, "sources": [ "mimo-v2-5-s1", "mimo-v2-5-s2" ] } ] } }, { "id": "mimo-v2-5-pro", "model": "MiMo-V2.5-Pro", "company": "Xiaomi", "family": "MiMo", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "agentic" ], "tags": [], "note": "Released alongside MiMo-V2.5: 1T total parameters, 42B activated and a 1M-token context (1,048,576 in the published config), aimed at high-intensity agent work.", "events": [ { "type": "announcement", "date": "2026-04-23", "sources": [ "mimo-v2-5-pro-s1", "mimo-v2-5-pro-s2" ] } ], "specifications": { "language": { "context_window": 1048576, "parameter_count": 1000000000000 } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "mimo-v2-5-pro-s1", "url": "https://mimo.mi.com/docs/en-US/updates/model", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260616040133/https://mimo.mi.com/docs/en-US/updates/model", "retrieved": "2026-06-16" }, { "id": "mimo-v2-5-pro-s2", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260427174214/https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro", "retrieved": "2026-04-27" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-04-23", "sources": [ "mimo-v2-5-pro-s1" ] } ], "context_window": [ { "value": 1048576, "sources": [ "mimo-v2-5-pro-s1", "mimo-v2-5-pro-s2" ] } ], "parameter_count": [ { "value": 1000000000000, "sources": [ "mimo-v2-5-pro-s1", "mimo-v2-5-pro-s2" ] } ] } }, { "id": "grok-voice-think-fast-1-0", "model": "Grok Voice Think Fast 1.0", "company": "xAI", "family": "Grok Voice", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "xAI made its voice agent model grok-voice-think-fast-1.0 available through the Speech to Speech API.", "events": [ { "type": "announcement", "date": "2026-04-23", "sources": [ "grok-voice-think-fast-1-0-s1", "grok-voice-think-fast-1-0-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "grok-voice-think-fast-1-0-s1", "url": "https://x.ai/news/grok-voice-think-fast-1", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260423223349/https://x.ai/news/grok-voice-think-fast-1", "retrieved": "2026-04-23" }, { "id": "grok-voice-think-fast-1-0-s2", "url": "https://docs.x.ai/developers/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260501083732/https://docs.x.ai/developers/release-notes", "retrieved": "2026-05-01" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-04-23", "sources": [ "grok-voice-think-fast-1-0-s1" ] } ] } }, { "id": "deepseek-v4-flash", "model": "DeepSeek V4 Flash", "company": "DeepSeek", "family": "DeepSeek", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding", "reasoning" ], "tags": [], "note": "A 284B MoE with 13B active parameters and 1M context, MIT-licensed alongside V4-Pro.", "events": [ { "type": "announcement", "date": "2026-04-24", "sources": [ "deepseek-v4-flash-s1", "deepseek-v4-flash-s2" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": 284000000000 } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "deepseek-v4-flash-s1", "url": "https://api-docs.deepseek.com/news/news260424", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260424071346/https://api-docs.deepseek.com/news/news260424", "retrieved": "2026-04-24" }, { "id": "deepseek-v4-flash-s2", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260424025422/https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash", "retrieved": "2026-04-24" }, { "id": "deepseek-v4-flash-price-20260602", "url": "https://api-docs.deepseek.com/quick_start/pricing/", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260602230941/https://api-docs.deepseek.com/quick_start/pricing/", "retrieved": "2026-06-02" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (2026-04-24), context window (1M) and parameter count (284B) — was found in an archived primary source, checked 2026-08-10. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present. Capabilities evidenced from the primary sources: coding: “co/collections/deepseek-ai/deepseek-v4 DeepSeek-V4-Pro ​ 🔹 Enhanced Agentic Capabilities: Open-source SOTA in Agentic Coding benchmarks.” | reasoning: “DeepSeek-V4-Pro-Max , the maximum reasoning effort mode of DeepSeek-V4-Pro, significantly advances the knowledge capabilities of open-source models, firmly establishing i”." }, "evidence": { "release_date": [ { "value": "2026-04-24", "sources": [ "deepseek-v4-flash-s2" ] } ], "context_window": [ { "value": 1000000, "sources": [ "deepseek-v4-flash-s1", "deepseek-v4-flash-s2" ] } ], "parameter_count": [ { "value": 284000000000, "sources": [ "deepseek-v4-flash-s1", "deepseek-v4-flash-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 0.14, "output": 0.28 }, "currency": "USD", "observed_on": "2026-06-02", "sources": [ "deepseek-v4-flash-price-20260602" ], "note": "Standard-hours cache-miss input and output as the page read on this date. Cache-hit input and any off-peak discount are priced separately and are not recorded here." } ] }, { "id": "hy3-preview", "model": "Hy3 preview", "company": "Tencent", "family": "Hy", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "agentic" ], "tags": [], "note": "Tencent launched and open-sourced Hy3 preview, a Mixture-of-Experts model integrating fast and slow thinking, with 295 billion total parameters, 21 billion activated and a context window of up to 256K tokens (262,144 in the published config). Tencent's GitHub and Hugging Face repositories were created on April 21, three days before the announcement.", "events": [ { "type": "announcement", "date": "2026-04-24", "sources": [ "hy3-preview-s1", "hy3-preview-s2" ] } ], "specifications": { "language": { "context_window": 262144, "parameter_count": 295000000000 } }, "access": { "open_weights": true, "license": null }, "sources": [ { "id": "hy3-preview-s1", "url": "https://www.tencent.com/tencent-unveils-hy3-preview-model-enhances-agent-capabilities-and-real-world-usability/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260716205033/https://www.tencent.com/tencent-unveils-hy3-preview-model-enhances-agent-capabilities-and-real-world-usability/", "retrieved": "2026-07-16" }, { "id": "hy3-preview-s2", "url": "https://huggingface.co/tencent/Hy3-preview", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260424043821/https://huggingface.co/tencent/Hy3-preview", "retrieved": "2026-04-24" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`. Modalities recorded as text-to-text. The primary source states the model is text-only." }, "evidence": { "release_date": [ { "value": "2026-04-24", "sources": [ "hy3-preview-s1" ] } ], "context_window": [ { "value": 262144, "sources": [ "hy3-preview-s1", "hy3-preview-s2" ] } ], "parameter_count": [ { "value": 295000000000, "sources": [ "hy3-preview-s1", "hy3-preview-s2" ] } ] } }, { "id": "gemini-3-1-flash-lite", "model": "Gemini 3.1 Flash-Lite", "company": "Google DeepMind", "family": "Gemini", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "function_calling", "structured_output" ], "tags": [], "note": "The lightweight text tier of the 3.1 generation, released to general availability. Distinct from gemini-3.1-flash-lite-image (Nano Banana 2 Lite), which shipped 2026-06-30.", "events": [ { "type": "announcement", "date": "2026-05-07", "sources": [ "gemini-3-1-flash-lite-s1" ] } ], "specifications": { "language": { "context_window": 1048576, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-3-1-flash-lite-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20250605063552/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2025-06-05" }, { "id": "gemini-3-1-flash-lite-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260507181345/https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite?hl=en", "retrieved": "2026-05-07" }, { "id": "gemini-3-1-flash-lite-price-20260602", "url": "https://ai.google.dev/gemini-api/docs/pricing", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260602081519/https://ai.google.dev/gemini-api/docs/pricing", "retrieved": "2026-06-02" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date and the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 0.25, "output": 1.5 }, "currency": "USD", "observed_on": "2026-06-02", "sources": [ "gemini-3-1-flash-lite-price-20260602" ], "note": "Paid-tier input and output as the page read on this date. The free tier and context-caching rates are priced separately and are not recorded here." } ] }, { "id": "qwen-3-7-max", "model": "Qwen3.7-Max", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "Alibaba's proprietary agent-focused flagship, served through Alibaba Cloud Model Studio.", "events": [ { "type": "announcement", "date": "2026-05-16", "sources": [ "qwen-3-7-max-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "qwen-3-7-max-s1", "url": "https://qwen.ai/blog?id=qwen3.7", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260520025139/https://qwen.ai/blog?id=qwen3.7", "retrieved": "2026-05-20" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present." }, "evidence": {}, "undisclosed": [ "context_window", "parameter_count" ] }, { "id": "gemini-3-5-flash", "model": "Gemini 3.5 Flash", "company": "Google DeepMind", "family": "Gemini", "classification": { "primary_type": "language", "subtype": "slm" }, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "capabilities": [ "agentic", "reasoning", "function_calling", "structured_output" ], "tags": [ "small-efficient" ], "note": "Google's agent-optimized workhorse model launched at I/O with 1M-token context.", "events": [ { "type": "announcement", "date": "2026-05-19", "sources": [ "gemini-3-5-flash-s1", "gemini-3-5-flash-s2" ] } ], "specifications": { "language": { "context_window": 1048576, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-3-5-flash-s1", "url": "https://en.wikipedia.org/wiki/Gemini_(language_model)", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20260516195051/https://en.wikipedia.org/wiki/Gemini_(language_model)", "retrieved": "2026-05-16" }, { "id": "gemini-3-5-flash-s2", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260519180605/https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash", "retrieved": "2026-05-19" }, { "id": "gemini-3-5-flash-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260525114658/https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash?hl=en", "retrieved": "2026-05-25" }, { "id": "gemini-3-5-flash-price-20260602", "url": "https://ai.google.dev/gemini-api/docs/pricing", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260602081519/https://ai.google.dev/gemini-api/docs/pricing", "retrieved": "2026-06-02" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities taken from the documentation's own declaration: \"Inputs Text, Image, Video, Audio, and PDF Output Text\"." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2026-05-19", "sources": [ "gemini-3-5-flash-gdocs" ] } ], "context_window": [ { "value": 1048576, "sources": [ "gemini-3-5-flash-s2", "gemini-3-5-flash-gdocs" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 1.5, "output": 9 }, "currency": "USD", "observed_on": "2026-06-02", "sources": [ "gemini-3-5-flash-price-20260602" ], "note": "Paid-tier input and output as the page read on this date. The free tier and context-caching rates are priced separately and are not recorded here." } ] }, { "id": "qwen-3-7-plus", "model": "Qwen3.7-Plus", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "The multimodal agent model of the Qwen3.7 line, released five days after Qwen3.7-Max.", "events": [ { "type": "announcement", "date": "2026-05-21", "sources": [ "qwen-3-7-plus-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "qwen-3-7-plus-s1", "url": "https://qwen.ai/blog?id=qwen3.7-plus", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260601182713/https://qwen.ai/blog?id=qwen3.7-plus", "retrieved": "2026-06-01" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present." }, "evidence": {}, "undisclosed": [ "context_window", "parameter_count" ] }, { "id": "hy-mt2-1-8b", "model": "Hy-MT2-1.8B", "company": "Tencent", "family": "Hy", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "multilingual" ], "tags": [], "note": "The smallest of three multilingual translation models Tencent open-sourced together as Hy-MT2 on 2026-05-21.", "events": [ { "type": "announcement", "date": "2026-05-21", "sources": [ "hy-mt2-1-8b-s1", "hy-mt2-1-8b-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "hy-mt2-1-8b-s1", "url": "https://github.com/Tencent-Hunyuan/Hy-MT2", "type": "official_repository", "authority": "primary", "archived_url": "https://web.archive.org/web/20260521084901/https://github.com/Tencent-Hunyuan/Hy-MT2", "retrieved": "2026-05-21" }, { "id": "hy-mt2-1-8b-s2", "url": "https://huggingface.co/tencent/Hy-MT2-1.8B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260521084916/https://huggingface.co/tencent/Hy-MT2-1.8B", "retrieved": "2026-05-21" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-05-21", "sources": [ "hy-mt2-1-8b-s2" ] } ] }, "undisclosed": [ "context_window", "parameter_count" ] }, { "id": "hy-mt2-7b", "model": "Hy-MT2-7B", "company": "Tencent", "family": "Hy", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "multilingual" ], "tags": [], "note": "One of three multilingual translation models Tencent open-sourced together as Hy-MT2 on 2026-05-21.", "events": [ { "type": "announcement", "date": "2026-05-21", "sources": [ "hy-mt2-7b-s1", "hy-mt2-7b-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "hy-mt2-7b-s1", "url": "https://github.com/Tencent-Hunyuan/Hy-MT2", "type": "official_repository", "authority": "primary", "archived_url": "https://web.archive.org/web/20260521084901/https://github.com/Tencent-Hunyuan/Hy-MT2", "retrieved": "2026-05-21" }, { "id": "hy-mt2-7b-s2", "url": "https://huggingface.co/tencent/Hy-MT2-7B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260521205706/https://huggingface.co/tencent/Hy-MT2-7B", "retrieved": "2026-05-21" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-05-21", "sources": [ "hy-mt2-7b-s2" ] } ] }, "undisclosed": [ "context_window", "parameter_count" ] }, { "id": "hy-mt2-30b-a3b", "model": "Hy-MT2-30B-A3B", "company": "Tencent", "family": "Hy", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "multilingual" ], "tags": [], "note": "The largest of three multilingual translation models Tencent open-sourced together as Hy-MT2 on 2026-05-21, a Mixture-of-Experts model by its name.", "events": [ { "type": "announcement", "date": "2026-05-21", "sources": [ "hy-mt2-30b-a3b-s1", "hy-mt2-30b-a3b-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "hy-mt2-30b-a3b-s1", "url": "https://github.com/Tencent-Hunyuan/Hy-MT2", "type": "official_repository", "authority": "primary", "archived_url": "https://web.archive.org/web/20260521084901/https://github.com/Tencent-Hunyuan/Hy-MT2", "retrieved": "2026-05-21" }, { "id": "hy-mt2-30b-a3b-s2", "url": "https://huggingface.co/tencent/Hy-MT2-30B-A3B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260521205628/https://huggingface.co/tencent/Hy-MT2-30B-A3B", "retrieved": "2026-05-21" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-05-21", "sources": [ "hy-mt2-30b-a3b-s2" ] } ] }, "undisclosed": [ "context_window", "parameter_count" ] }, { "id": "mistral-medium-3-5", "model": "Mistral Medium 3.5", "company": "Mistral AI", "family": "Mistral", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "coding" ], "tags": [], "note": "A dense 128B open-weight model aimed at long-horizon coding and productivity work, released under a modified MIT licence.", "events": [ { "type": "announcement", "date": "2026-05-22", "sources": [ "mistral-medium-3-5-s1" ] } ], "specifications": { "language": { "context_window": 256000, "parameter_count": 128000000000 } }, "access": { "open_weights": true, "license": "Modified MIT" }, "sources": [ { "id": "mistral-medium-3-5-s1", "url": "https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260521023218/https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5", "retrieved": "2026-05-21" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement. Not verified because the release date could not be traced: every archived snapshot of those sources predates the release — the newest is 2026-05-21, captured before the page announced it — so the archived copy does not contain the figure. A capture from on or after 2026-05-22 would settle it." }, "evidence": { "context_window": [ { "value": 256000, "sources": [ "mistral-medium-3-5-s1" ] } ], "parameter_count": [ { "value": 128000000000, "sources": [ "mistral-medium-3-5-s1" ] } ] }, "benchmarks": [ { "name": "SWE-bench Verified", "score": 77.6, "evaluation_type": "vendor_reported", "reported_on": "2026-05-22", "sources": [ "mistral-medium-3-5-s1" ] } ], "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 1.5, "output": 7.5 }, "currency": "USD", "observed_on": "2026-05-21", "sources": [ "mistral-medium-3-5-s1" ] } ] }, { "id": "claude-opus-4-8", "model": "Claude Opus 4.8", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "agentic", "vision", "multilingual" ], "tags": [ "flagship" ], "note": "The final Opus 4.x release before the Claude 5 generation.", "events": [ { "type": "announcement", "date": "2026-05-28", "sources": [ "claude-opus-4-8-s1", "claude-opus-4-8-s2" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-opus-4-8-s1", "url": "https://www.anthropic.com/news/claude-opus-4-8", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260528165611/https://www.anthropic.com/news/claude-opus-4-8", "retrieved": "2026-05-28" }, { "id": "claude-opus-4-8-s2", "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260528070430/https://platform.claude.com/docs/en/about-claude/models/overview", "retrieved": "2026-05-28" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (May 28, 2026) and context window (1M) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Capabilities evidenced from the primary sources: agentic: “It is our most capable generally available model, with a step-change improvement in agentic coding over Claude Opus 4.” | multilingual: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.” | vision: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2026-05-28", "sources": [ "claude-opus-4-8-s1" ] } ], "context_window": [ { "value": 1000000, "sources": [ "claude-opus-4-8-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 5, "output": 25 }, "currency": "USD", "sources": [ "claude-opus-4-8-s1" ], "observed_on": "2026-05-28" } ] }, { "id": "nano-banana-pro", "model": "Nano Banana Pro", "company": "Google DeepMind", "family": "Nano Banana", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": { "input": [ "image", "text" ], "output": [ "image", "text" ] }, "capabilities": [ "reasoning", "image_generation" ], "tags": [], "note": "The Pro tier of Google's native image model, served as gemini-3-pro-image.", "events": [ { "type": "announcement", "date": "2026-05-28", "sources": [ "nano-banana-pro-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "nano-banana-pro-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20250605063552/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2025-06-05" }, { "id": "nano-banana-pro-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3-pro-image?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260608132629/https://ai.google.dev/gemini-api/docs/models/gemini-3-pro-image?hl=en", "retrieved": "2026-06-08" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." } }, { "id": "grok-build-0-1", "model": "Grok Build 0.1", "company": "xAI", "family": "Grok Build", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "coding", "agentic", "function_calling", "structured_output", "reasoning" ], "tags": [], "note": "xAI's coding model trained for agentic coding workflows, made available on the xAI API as grok-build-0.1 after early access. It is the model; Grok Build, the coding agent it powers, is recorded as a milestone.", "events": [ { "type": "announcement", "date": "2026-05-29", "sources": [ "grok-build-0-1-s1", "grok-build-0-1-s2" ] } ], "specifications": { "language": { "context_window": 256000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "grok-build-0-1-s1", "url": "https://x.ai/news/grok-build-0-1", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260530201923/https://x.ai/news/grok-build-0-1", "retrieved": "2026-05-30" }, { "id": "grok-build-0-1-s2", "url": "https://docs.x.ai/developers/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260530201148/https://docs.x.ai/developers/release-notes", "retrieved": "2026-05-30" }, { "id": "grok-build-0-1-xdocs", "url": "https://docs.x.ai/developers/models.md", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null }, { "id": "grok-build-0-1-xmodel", "url": "https://docs.x.ai/developers/models/grok-build-0.1", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "partially_verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`. Downgraded: context_window comes from the lab's documentation, but attribute-facts could not read an archived snapshot that states it, so the figure is not yet traced to a quoted sentence. reconcile-status upgrades it once attribute-facts succeeds." }, "evidence": { "release_date": [ { "value": "2026-05-29", "sources": [ "grok-build-0-1-s1" ] } ] } }, { "id": "ornith-1-0-9b", "model": "Ornith-1.0-9B", "company": "Ornith", "family": "Ornith", "classification": { "primary_type": "language", "subtype": "reasoning" }, "modalities": null, "capabilities": [ "coding", "agentic", "reasoning" ], "tags": [ "small-efficient" ], "note": "The compact dense member of Ornith-1.0, post-trained on Gemma 4 and Qwen 3.5 with a self-scaffolding RL framework and published for edge deployment.", "events": [ { "type": "announcement", "date": "2026-06", "sources": [ "ornith-1-0-9b-s1", "ornith-1-0-9b-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 9000000000 } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "ornith-1-0-9b-s1", "url": "https://ornith.ai/ornith_1_0.html", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260807012121/https://ornith.ai/ornith_1_0.html", "retrieved": "2026-08-07" }, { "id": "ornith-1-0-9b-s2", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-9B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260807133215/https://huggingface.co/ornith-ai/Ornith-1.0-9B", "retrieved": "2026-08-07" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-06", "sources": [ "ornith-1-0-9b-s2" ] } ], "parameter_count": [ { "value": 9000000000, "sources": [ "ornith-1-0-9b-s1", "ornith-1-0-9b-s2" ] } ] } }, { "id": "ornith-1-0-35b", "model": "Ornith-1.0-35B", "company": "Ornith", "family": "Ornith", "classification": { "primary_type": "language", "subtype": "reasoning" }, "modalities": null, "capabilities": [ "coding", "agentic", "reasoning" ], "tags": [], "note": "The mid-size mixture-of-experts member of Ornith-1.0, reported to outperform Qwen 3.5-397B on Terminal-Bench 2.1 at a fraction of the size.", "events": [ { "type": "announcement", "date": "2026-06", "sources": [ "ornith-1-0-35b-s1", "ornith-1-0-35b-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 35000000000 } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "ornith-1-0-35b-s1", "url": "https://ornith.ai/ornith_1_0.html", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260807012121/https://ornith.ai/ornith_1_0.html", "retrieved": "2026-08-07" }, { "id": "ornith-1-0-35b-s2", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-35B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260807133131/https://huggingface.co/ornith-ai/Ornith-1.0-35B", "retrieved": "2026-08-07" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-06", "sources": [ "ornith-1-0-35b-s2" ] } ], "parameter_count": [ { "value": 35000000000, "sources": [ "ornith-1-0-35b-s1", "ornith-1-0-35b-s2" ] } ] } }, { "id": "ornith-1-0-397b", "model": "Ornith-1.0-397B", "company": "Ornith", "family": "Ornith", "classification": { "primary_type": "language", "subtype": "reasoning" }, "modalities": null, "capabilities": [ "coding", "agentic", "reasoning" ], "tags": [ "flagship" ], "note": "The frontier-scale mixture-of-experts member of Ornith-1.0, and the model the launch was headlined on.", "events": [ { "type": "announcement", "date": "2026-06", "sources": [ "ornith-1-0-397b-s1", "ornith-1-0-397b-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 397000000000 } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "ornith-1-0-397b-s1", "url": "https://ornith.ai/ornith_1_0.html", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260807012121/https://ornith.ai/ornith_1_0.html", "retrieved": "2026-08-07" }, { "id": "ornith-1-0-397b-s2", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-397B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260821051951/https://huggingface.co/ornith-ai/Ornith-1.0-397B", "retrieved": "2026-08-21" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-06", "sources": [ "ornith-1-0-397b-s2" ] } ], "parameter_count": [ { "value": 397000000000, "sources": [ "ornith-1-0-397b-s1", "ornith-1-0-397b-s2" ] } ] } }, { "id": "minimax-m3", "model": "MiniMax M3", "company": "MiniMax", "family": "MiniMax", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision" ], "tags": [], "note": "MiniMax's multimodal language model, taking image and text input to produce text.", "events": [ { "type": "announcement", "date": "2026-06-02", "sources": [ "minimax-m3-s1", "minimax-m3-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": true, "license": "minimax-community" }, "sources": [ { "id": "minimax-m3-s1", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M3", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260612193019/https://huggingface.co/MiniMaxAI/MiniMax-M3", "retrieved": "2026-06-12" }, { "id": "minimax-m3-s2", "url": "https://www.minimax.io/blog/minimax-m3", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260602105413/https://www.minimax.io/blog/minimax-m3", "retrieved": "2026-06-02" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the official model card and the lab's own announcement, and every value it asserts was found there. Licence from the model card's own license_name field on MiniMaxAI/MiniMax-M3." }, "evidence": { "release_date": [ { "value": "2026-06-02", "sources": [ "minimax-m3-s1" ] } ] }, "undisclosed": [ "context_window" ] }, { "id": "mai-image-2-5", "model": "MAI-Image-2.5", "company": "Microsoft", "family": "MAI-Image", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "Image generation and editing model from Microsoft's Superintelligence team.", "events": [ { "type": "announcement", "date": "2026-06-02", "sources": [ "mai-image-2-5-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mai-image-2-5-s1", "url": "https://microsoft.ai/news/introducing-mai-image-2-5/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260602211645/https://microsoft.ai/news/introducing-mai-image-2-5/", "retrieved": "2026-06-02" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2026-06-02", "sources": [ "mai-image-2-5-s1" ] } ] } }, { "id": "mai-voice-2", "model": "MAI-Voice-2", "company": "Microsoft", "family": "MAI-Voice", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [ "speech_generation" ], "tags": [], "note": "Multilingual text-to-speech with voice cloning.", "events": [ { "type": "announcement", "date": "2026-06-02", "sources": [ "mai-voice-2-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mai-voice-2-s1", "url": "https://microsoft.ai/news/mai-voice-2/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260712151543/http://microsoft.ai/news/mai-voice-2/", "retrieved": "2026-07-12" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there. Capabilities evidenced from the primary sources: speech_generation: “com LinkedIn Instagram X Off Accessibility Mode Off Accessibility Mode Copilot --> Models Introducing MAI-Voice-2 Superintelligence team June 2, 2026 Models Superintelligence team ”." }, "evidence": { "release_date": [ { "value": "2026-06-02", "sources": [ "mai-voice-2-s1" ] } ] } }, { "id": "mai-transcribe-1-5", "model": "MAI-Transcribe-1.5", "company": "Microsoft", "family": "MAI-Transcribe", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "Speech-to-text model, positioned for production transcription workloads.", "events": [ { "type": "announcement", "date": "2026-06-02", "sources": [ "mai-transcribe-1-5-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mai-transcribe-1-5-s1", "url": "https://microsoft.ai/news/mai-transcribe-1-5more-accurate-context-aware-and-built-for-production/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260603031322/https://microsoft.ai/news/mai-transcribe-1-5more-accurate-context-aware-and-built-for-production/", "retrieved": "2026-06-03" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2026-06-02", "sources": [ "mai-transcribe-1-5-s1" ] } ] } }, { "id": "mai-code-1-flash", "model": "MAI-Code-1-Flash", "company": "Microsoft", "family": "MAI", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "coding" ], "tags": [], "note": "Lightweight coding model for agentic developer workflows.", "events": [ { "type": "announcement", "date": "2026-06-02", "sources": [ "mai-code-1-flash-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mai-code-1-flash-s1", "url": "https://microsoft.ai/news/introducingmai-code-1-flash/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260602191242/https://microsoft.ai/news/introducingmai-code-1-flash/", "retrieved": "2026-06-02" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2026-06-02", "sources": [ "mai-code-1-flash-s1" ] } ] }, "undisclosed": [ "context_window", "parameter_count" ] }, { "id": "gemma-4-12b", "model": "Gemma 4 12B", "company": "Google DeepMind", "family": "Gemma", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image", "audio" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "agentic", "vision", "coding" ], "tags": [], "note": "Sits between Gemma 4 E4B and the 26B MoE, and is Google's first mid-sized Gemma with native audio input. Drops the separate multimodal encoders: vision goes through a lightweight embedding module and raw audio is projected straight into the text token space, so both flow into the LLM backbone directly.", "events": [ { "type": "announcement", "date": "2026-06-03", "sources": [ "gemma-4-12b-s1", "gemma-4-12b-s2" ] }, { "type": "weights_availability", "date": "2026-06-03", "sources": [ "gemma-4-12b-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 12000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "gemma-4-12b-s1", "url": "https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12b/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260603163709/https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12b/", "retrieved": "2026-06-03" }, { "id": "gemma-4-12b-s2", "url": "https://huggingface.co/google/gemma-4-12B-it", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260603160621/https://huggingface.co/google/gemma-4-12B-it", "retrieved": "2026-06-03" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement and the official model card. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities read from the launch announcement, which states: “Novel unified architecture: No multimodal encoders. The vision and audio inputs flow directly into the LLM backbone,” and describes each path separately — vision through a lightweight embedding module, raw audio projected into the same dimensional space as text tokens. Output is recorded as text: no image, audio or video generation is claimed anywhere in the primary sources. Capabilities evidenced from the primary sources: agentic: “5X 2X Today, we are introducing Gemma 4 12B, our latest model designed to bring agentic multimodal intelligence directly to laptops.” | vision: “Gemma 4 models are multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output.” | coding: “Featuring both Dense and Mixture-of-Experts (MoE) architectures, Gemma 4 is well-suited for tasks like text generation, coding, and reasoning.” | reasoning: “Thinking Mode Configuration Compared to Gemma 3, the models use standard system , assistant , and user roles.”." }, "evidence": { "release_date": [ { "value": "2026-06-03", "sources": [ "gemma-4-12b-s1" ] } ], "parameter_count": [ { "value": 12000000000, "sources": [ "gemma-4-12b-s1" ] } ] }, "benchmarks": [ { "name": "LiveCodeBench", "score": 80, "evaluation_type": "vendor_reported", "reported_on": "2026-06-03", "sources": [ "gemma-4-12b-s2" ] }, { "name": "MMMU", "score": 76.9, "evaluation_type": "vendor_reported", "reported_on": "2026-06-03", "sources": [ "gemma-4-12b-s2" ] }, { "name": "GPQA Diamond", "score": 84.3, "evaluation_type": "vendor_reported", "reported_on": "2026-06-03", "sources": [ "gemma-4-12b-s2" ] }, { "name": "AIME", "score": 89.2, "evaluation_type": "vendor_reported", "reported_on": "2026-06-03", "sources": [ "gemma-4-12b-s2" ] }, { "name": "MMLU", "score": 85.2, "evaluation_type": "vendor_reported", "reported_on": "2026-06-03", "sources": [ "gemma-4-12b-s2" ] } ] }, { "id": "grok-imagine-video-1-5-preview", "model": "Grok Imagine Video 1.5 Preview", "company": "xAI", "family": "Grok Imagine", "classification": { "primary_type": "video_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "xAI released grok-imagine-video-1.5-preview as its latest image-to-video model, thirteen days before Grok Imagine Video 1.5.", "events": [ { "type": "announcement", "date": "2026-06-03", "sources": [ "grok-imagine-video-1-5-preview-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "grok-imagine-video-1-5-preview-s1", "url": "https://x.ai/news/grok-imagine-1-5", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260606151713/https://x.ai/news/grok-imagine-1-5", "retrieved": "2026-06-06" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-06-03", "sources": [ "grok-imagine-video-1-5-preview-s1" ] } ] } }, { "id": "nemotron-3-ultra", "model": "Nemotron 3 Ultra", "company": "NVIDIA", "family": "Nemotron", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "agentic" ], "tags": [], "note": "A 550B open MoE hybrid Mamba-Transformer model built for long-running agents.", "events": [ { "type": "announcement", "date": "2026-06-04", "sources": [ "nemotron-3-ultra-s1", "nemotron-3-ultra-s2" ] }, { "type": "weights_availability", "date": "2026-06-04", "sources": [ "nemotron-3-ultra-s1" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": 550000000000 } }, "access": { "open_weights": true, "license": "OpenMDW License 1.1" }, "sources": [ { "id": "nemotron-3-ultra-s1", "url": "https://research.nvidia.com/labs/nemotron/Nemotron-3-Ultra/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260604204450/https://research.nvidia.com/labs/nemotron/Nemotron-3-Ultra/", "retrieved": "2026-06-04" }, { "id": "nemotron-3-ultra-s2", "url": "https://openmdw.ai/license/1-1/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260528163158/https://openmdw.ai/license/1-1/", "retrieved": "2026-05-28" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present. Capabilities evidenced from the primary sources: agentic: “Nemotron-Posttraining-v3 : A collection of post-training datasets for improving agentic, reasoning, and general model capabilities during SFT and RL.”." }, "evidence": { "release_date": [ { "value": "2026-06-04", "sources": [ "nemotron-3-ultra-s1" ] } ], "context_window": [ { "value": 1000000, "sources": [ "nemotron-3-ultra-s1" ] } ], "parameter_count": [ { "value": 550000000000, "sources": [ "nemotron-3-ultra-s1" ] } ] } }, { "id": "claude-fable-5", "model": "Claude Fable 5", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision", "multilingual", "reasoning", "agentic" ], "tags": [ "flagship" ], "note": "Opens the Claude 5 generation. Shares its underlying model with Mythos 5, with additional safeguards for biology, cybersecurity and LLM R&D.", "events": [ { "type": "announcement", "date": "2026-06-09", "sources": [ "claude-fable-5-s1", "claude-fable-5-s2" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-fable-5-s1", "url": "https://en.wikipedia.org/wiki/Claude_(language_model)", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20260609152642/https://en.wikipedia.org/wiki/Claude_(language_model)", "retrieved": "2026-06-09" }, { "id": "claude-fable-5-s2", "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260608180822/https://platform.claude.com/docs/en/about-claude/models/overview", "retrieved": "2026-06-08" }, { "id": "claude-fable-5-adocs", "url": "https://platform.claude.com/docs/en/about-claude/models/overview.md", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260817055519/https://platform.claude.com/docs/en/about-claude/models/overview.md", "retrieved": "2026-08-17" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Capabilities evidenced from the primary sources: multilingual: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.” | vision: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2026-06-09", "sources": [ "claude-fable-5-adocs" ] } ], "context_window": [ { "value": 1000000, "sources": [ "claude-fable-5-s2", "claude-fable-5-adocs" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 10, "output": 50 }, "currency": "USD", "observed_on": "2026-08-17", "sources": [ "claude-fable-5-adocs" ], "note": "Standard tier. Anthropic prices batch and cached tokens differently." } ] }, { "id": "claude-mythos-5", "model": "Claude Mythos 5", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "vision", "multilingual" ], "tags": [ "flagship" ], "note": "Released the same day as Fable 5 with limited availability through Project Glasswing.", "events": [ { "type": "announcement", "date": "2026-06-09", "sources": [ "claude-mythos-5-s1", "claude-mythos-5-s2" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-mythos-5-s1", "url": "https://en.wikipedia.org/wiki/Claude_(language_model)", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20260609152642/https://en.wikipedia.org/wiki/Claude_(language_model)", "retrieved": "2026-06-09" }, { "id": "claude-mythos-5-s2", "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260608180822/https://platform.claude.com/docs/en/about-claude/models/overview", "retrieved": "2026-06-08" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Capabilities evidenced from the primary sources: multilingual: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.” | vision: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.”." }, "undisclosed": [ "parameter_count" ], "evidence": { "context_window": [ { "value": 1000000, "sources": [ "claude-mythos-5-s2" ] } ] } }, { "id": "glm-5-2", "model": "GLM-5.2", "company": "Zhipu AI", "family": "GLM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "agentic", "coding", "reasoning", "tool_use" ], "tags": [], "note": "A 744B MIT-licensed open model, the first open model to top closed frontiers on agentic coding.", "events": [ { "type": "announcement", "date": "2026-06-13", "sources": [ "glm-5-2-s1", "glm-5-2-s2" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": 753000000000 } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "glm-5-2-s1", "url": "https://z.ai/blog/glm-5.2", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260616175106/https://z.ai/blog/glm-5.2", "retrieved": "2026-06-16" }, { "id": "glm-5-2-s2", "url": "https://huggingface.co/zai-org/GLM-5.2", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260616174532/https://huggingface.co/zai-org/GLM-5.2", "retrieved": "2026-06-16" } ], "provenance": { "status": "partially_verified", "confidence": 80, "reason": "Cited to the lab's own announcement and the official model card. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Zhipu's own release notes date GLM-5.2 to 2026-06-16 where this record publishes 2026-06-13; the disagreement is recorded rather than resolved." }, "evidence": { "context_window": [ { "value": 1000000, "sources": [ "glm-5-2-s2" ] } ], "parameter_count": [ { "value": 753000000000, "sources": [ "glm-5-2-s2" ] } ] } }, { "id": "grok-imagine-video-1-5", "model": "Grok Imagine Video 1.5", "company": "xAI", "family": "Grok Imagine", "classification": { "primary_type": "video_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "xAI's video generation model grok-imagine-video-1.5, with improved quality at faster speeds than the preview. On 2026-07-31 it gained text, image and voice references; xAI's release notes add native 1080p.", "events": [ { "type": "announcement", "date": "2026-06-16", "sources": [ "grok-imagine-video-1-5-s1", "grok-imagine-video-1-5-s2" ] }, { "type": "major_update", "date": "2026-07-31", "sources": [ "grok-imagine-video-1-5-s3" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "grok-imagine-video-1-5-s1", "url": "https://x.ai/news/grok-imagine-video-1-5", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260618200828/https://x.ai/news/grok-imagine-video-1-5", "retrieved": "2026-06-18" }, { "id": "grok-imagine-video-1-5-s2", "url": "https://docs.x.ai/developers/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260616000754/https://docs.x.ai/developers/release-notes", "retrieved": "2026-06-16" }, { "id": "grok-imagine-video-1-5-s3", "url": "https://x.ai/news/grok-imagine-video-1-5-references", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260801013202/https://x.ai/news/grok-imagine-video-1-5-references", "retrieved": "2026-08-01" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-06-16", "sources": [ "grok-imagine-video-1-5-s1" ] } ] } }, { "id": "mistral-ocr-4", "model": "Mistral OCR 4", "company": "Mistral AI", "family": "Mistral OCR", "classification": { "primary_type": "vision", "subtype": null }, "modalities": null, "capabilities": [ "coding", "agentic" ], "tags": [], "note": "Mistral's document-understanding model, reading page images into structured text.", "events": [ { "type": "announcement", "date": "2026-06-23", "sources": [ "mistral-ocr-4-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mistral-ocr-4-s1", "url": "https://mistral.ai/news/ocr-4", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260623144429/https://mistral.ai/news/ocr-4/", "retrieved": "2026-06-23" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there. Capabilities evidenced from the primary sources: coding: “Pricing Plans API pricing For enterprises Services Delivery methodology Model customization Industries Financial services Public sector & government Manufacturing Use cases Use cas” | agentic: “Benchmarks “We benchmarked Mistral OCR 4 against the leading agentic document parsers across a chart and figure dense financial QA dataset and reached equivalent accuracy at roughl”." }, "evidence": { "release_date": [ { "value": "2026-06-23", "sources": [ "mistral-ocr-4-s1" ] } ] } }, { "id": "seed-2-1", "model": "Seed2.1", "company": "ByteDance", "family": "Seed", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [], "tags": [], "note": "ByteDance's agent model for real-world productivity, and the flagship of the Seed line.", "events": [ { "type": "announcement", "date": "2026-06-23", "sources": [ "seed-2-1-s1", "seed-2-1-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "seed-2-1-s1", "url": "https://seed.bytedance.com/blog/seed2-1-officially-released-advancing-ai-productivity", "type": "official_announcement", "authority": "primary", "archived_url": null, "retrieved": null }, { "id": "seed-2-1-s2", "url": "https://seed.bytedance.com/en/seed2_1", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260623045712/https://seed.bytedance.com/en/seed2_1", "retrieved": "2026-06-23" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement and official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": {} }, { "id": "claude-sonnet-5", "model": "Claude Sonnet 5", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "agentic", "vision", "multilingual", "reasoning" ], "tags": [], "note": "The most agentic Sonnet to date, approaching Opus-class performance at a lower price.", "events": [ { "type": "announcement", "date": "2026-06-30", "sources": [ "claude-sonnet-5-s1", "claude-sonnet-5-s2" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-sonnet-5-s1", "url": "https://www.anthropic.com/news/claude-sonnet-5", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260630175737/https://www.anthropic.com/news/claude-sonnet-5", "retrieved": "2026-06-30" }, { "id": "claude-sonnet-5-s2", "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260630015858/https://platform.claude.com/docs/en/about-claude/models/overview", "retrieved": "2026-06-30" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (Jun 30, 2026) and context window (1M) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Capabilities evidenced from the primary sources: agentic: “Introducing Claude Sonnet 5 \\ Anthropic Skip to main content Skip to footer Research Policy Commitments Learn News Try Claude Product Introducing Claude Sonnet 5 Jun 30, ” | multilingual: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.” | vision: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2026-06-30", "sources": [ "claude-sonnet-5-s1" ] } ], "context_window": [ { "value": 1000000, "sources": [ "claude-sonnet-5-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 2, "output": 10 }, "currency": "USD", "sources": [ "claude-sonnet-5-s1" ], "observed_on": "2026-06-30" } ] }, { "id": "gemini-omni-flash", "model": "Gemini Omni Flash", "company": "Google DeepMind", "family": "Gemini Omni", "classification": { "primary_type": "video_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "A multimodal model for high-speed video generation and conversational video editing, released in public preview.", "events": [ { "type": "announcement", "date": "2026-06-30", "sources": [ "gemini-omni-flash-s1", "gemini-omni-flash-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-omni-flash-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260701184207/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2026-07-01" }, { "id": "gemini-omni-flash-s2", "url": "https://deepmind.google/models/gemini-omni/", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20260519193103/https://deepmind.google/models/gemini-omni/", "retrieved": "2026-05-19" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." } }, { "id": "nano-banana-2-lite", "model": "Nano Banana 2 Lite", "company": "Google DeepMind", "family": "Nano Banana", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": { "input": [ "text", "image" ], "output": [ "image", "text" ] }, "capabilities": [ "reasoning", "function_calling", "image_generation" ], "tags": [], "note": "The low-latency image generation and editing tier, served as gemini-3.1-flash-lite-image.", "events": [ { "type": "announcement", "date": "2026-06-30", "sources": [ "nano-banana-2-lite-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "nano-banana-2-lite-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20250605063552/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2025-06-05" }, { "id": "nano-banana-2-lite-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite-image?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260704003755/https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite-image?hl=en", "retrieved": "2026-07-04" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." } }, { "id": "hy3", "model": "Hy3", "company": "Tencent", "family": "Hy", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "reasoning", "agentic" ], "tags": [], "note": "Tencent officially launched Hy3, building on Hy3 preview with better performance, stability and cost efficiency: 295 billion total parameters, 21 billion active, and up to 256K context (262,144 in the published config). Tencent made it available globally on 2026-08-05. Hy is the brand formerly known as Hunyuan.", "events": [ { "type": "announcement", "date": "2026-07-06", "sources": [ "hy3-s1", "hy3-s2", "hy3-s3" ] }, { "type": "public_availability", "date": "2026-08-05", "sources": [ "hy3-s2" ] } ], "specifications": { "language": { "context_window": 262144, "parameter_count": 295000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "hy3-s1", "url": "https://www.tencent.com/en-us/articles/2202386.html", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260706191804/https://www.tencent.com/en-us/articles/2202386.html", "retrieved": "2026-07-06" }, { "id": "hy3-s2", "url": "https://www.tencent.com/tencent-hy3-now-available-globally-extending-practical-ai-across-products-workflows-and-cloud-services/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260828145810/https://www.tencent.com/tencent-hy3-now-available-globally-extending-practical-ai-across-products-workflows-and-cloud-services/", "retrieved": "2026-08-28" }, { "id": "hy3-s3", "url": "https://huggingface.co/tencent/Hy3", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260706083828/https://huggingface.co/tencent/Hy3", "retrieved": "2026-07-06" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-07-06", "sources": [ "hy3-s1", "hy3-s3" ] } ], "context_window": [ { "value": 262144, "sources": [ "hy3-s1", "hy3-s2", "hy3-s3" ] } ], "parameter_count": [ { "value": 295000000000, "sources": [ "hy3-s1", "hy3-s2", "hy3-s3" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 0.1288, "output": 0.5336 }, "currency": "USD", "observed_on": "2026-08-28", "sources": [ "hy3-s2" ] } ] }, { "id": "muse-image", "model": "Muse Image", "company": "Meta AI", "family": "Muse Image", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": null, "capabilities": [ "image_generation", "reasoning", "coding", "agentic", "tool_use" ], "tags": [], "note": "Meta Superintelligence Labs' first image generation model, launched in the Meta AI app, meta.ai, Instagram Stories in the US and WhatsApp in some countries. It works as an agent, calling search and coding tools and refining its own output. Announced alongside a preview of Muse Video, which was not yet available.", "events": [ { "type": "announcement", "date": "2026-07-07", "sources": [ "muse-image-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "muse-image-s1", "url": "https://ai.meta.com/blog/introducing-muse-image-muse-video-msl/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260707210624/https://ai.meta.com/blog/introducing-muse-image-muse-video-msl/", "retrieved": "2026-07-07" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`. Capabilities evidenced from the primary sources: image_generation: “Muse Image is our most advanced image generation model yet: it follows instructions faithfully, edits with precision, composes from multiple references, and draws on Instagram for ” | reasoning: “Muse Image: Agentic Image Generation Instead of directly mapping prompts to images, Muse Image operates as an agent: it invokes search and coding tools to improve accuracy, self-re” | coding: “Muse Image: Agentic Image Generation Instead of directly mapping prompts to images, Muse Image operates as an agent: it invokes search and coding tools to improve accuracy, self-re” | agentic: “Muse Image: Agentic Image Generation Instead of directly mapping prompts to images, Muse Image operates as an agent: it invokes search and coding tools to improve accuracy, self-re” | tool_use: “Tool Use We provide Muse Image with access to tools to enhance its agentic capabilities.”." }, "evidence": { "release_date": [ { "value": "2026-07-07", "sources": [ "muse-image-s1" ] } ] } }, { "id": "grok-4-5", "model": "Grok 4.5", "company": "xAI", "family": "Grok", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "function_calling", "structured_output", "coding", "agentic" ], "tags": [ "flagship" ], "note": "xAI's smartest model, its first release after going public and acquiring Cursor. xAI later re-dated its announcement page: the capture archived on 2026-07-08 reads \"Jul 8, 2026\", and the live page now reads \"Jul 16, 2026\". A page captured on July 8 was public on July 8, so that date stands; the later one likely reflects wider rollout.", "events": [ { "type": "announcement", "date": "2026-07-08", "sources": [ "grok-4-5-s1", "grok-4-5-s2" ] } ], "specifications": { "language": { "context_window": 500000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "grok-4-5-s1", "url": "https://x.ai/news/grok-4-5", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260708180848/https://x.ai/news/grok-4-5", "retrieved": "2026-07-08" }, { "id": "grok-4-5-s2", "url": "https://docs.x.ai/docs/models", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260204041847/https://docs.x.ai/docs/models", "retrieved": "2026-02-04" }, { "id": "grok-4-5-xmodel", "url": "https://docs.x.ai/developers/models/grok-4.5", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260710125453/https://docs.x.ai/developers/models/grok-4.5", "retrieved": "2026-07-10" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement and official documentation. Not verified because the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2026-07-08", "sources": [ "grok-4-5-s1" ] } ], "context_window": [ { "value": 500000, "sources": [ "grok-4-5-xmodel" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 2, "output": 6 }, "currency": "USD", "sources": [ "grok-4-5-s1" ], "observed_on": "2026-07-08" } ] }, { "id": "seedream-5-0-pro", "model": "Seedream 5.0 Pro", "company": "ByteDance", "family": "Seedream", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "A multimodal image generation model positioned around understanding design intent rather than generation alone.", "events": [ { "type": "announcement", "date": "2026-07-08", "sources": [ "seedream-5-0-pro-s1", "seedream-5-0-pro-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "seedream-5-0-pro-s1", "url": "https://seed.bytedance.com/blog/beyond-generation-it-understands-design-introducing-seedream-5-0-pro", "type": "official_announcement", "authority": "primary", "archived_url": null, "retrieved": null }, { "id": "seedream-5-0-pro-s2", "url": "https://seed.bytedance.com/en/seedream5_0_pro", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260708133912/https://seed.bytedance.com/en/seedream5_0_pro", "retrieved": "2026-07-08" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement and official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": {} }, { "id": "gpt-5-6-sol", "model": "GPT-5.6 Sol", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "agentic", "structured_output", "function_calling" ], "tags": [ "flagship" ], "note": "The frontier variant of the GPT-5.6 series.", "events": [ { "type": "announcement", "date": "2026-07-09", "sources": [ "gpt-5-6-sol-s1", "gpt-5-6-sol-s2" ] } ], "specifications": { "language": { "context_window": 1050000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-5-6-sol-s1", "url": "https://en.wikipedia.org/wiki/GPT-5.6", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20260709154055/https://en.wikipedia.org/wiki/GPT-5.6", "retrieved": "2026-07-09" }, { "id": "gpt-5-6-sol-s2", "url": "https://developers.openai.com/api/docs/models/gpt-5.6-sol", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260711182342/https://developers.openai.com/api/docs/models/gpt-5.6-sol", "retrieved": "2026-07-11" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Split from the former \"gpt-5-6\" record, which was named for the family while holding one variant's figures. Evidence: \"GPT-5.6 Sol is the frontier model in the GPT-5.6 series.\" Modalities taken from the documentation's own declaration: \"Input Text, image Output Text\". Capabilities evidenced from the primary sources: reasoning: “Reasoning token support”." }, "evidence": { "context_window": [ { "value": 1050000, "sources": [ "gpt-5-6-sol-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 5, "output": 30 }, "currency": "USD", "observed_on": "2026-08-17", "sources": [ "gpt-5-6-sol-s2" ], "note": "Base rate. Prompts above 272K tokens are billed at 2x input and 1.5x output." } ], "undisclosed": [ "parameter_count" ] }, { "id": "gpt-5-6-luna", "model": "GPT-5.6 Luna", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "agentic", "structured_output", "function_calling" ], "tags": [], "note": "The default variant of the GPT-5.6 series.", "events": [ { "type": "announcement", "date": "2026-07-09", "sources": [ "gpt-5-6-luna-s1", "gpt-5-6-luna-s2" ] } ], "specifications": { "language": { "context_window": 1050000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-5-6-luna-s1", "url": "https://en.wikipedia.org/wiki/GPT-5.6", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20260709154055/https://en.wikipedia.org/wiki/GPT-5.6", "retrieved": "2026-07-09" }, { "id": "gpt-5-6-luna-s2", "url": "https://developers.openai.com/api/docs/models/gpt-5.6-luna", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260713134652/https://developers.openai.com/api/docs/models/gpt-5.6-luna", "retrieved": "2026-07-13" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Split from the former \"gpt-5-6\" record, which was named for the family while holding one variant's figures. Evidence: \"GPT-5.6 Luna Default — its own model page in OpenAI's API documentation, with its own specifications.\" Modalities taken from the documentation's own declaration: \"Input Text, image Output Text\". Capabilities evidenced from the primary sources: reasoning: “Reasoning token support”." }, "evidence": { "context_window": [ { "value": 1050000, "sources": [ "gpt-5-6-luna-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 0.2, "output": 1.2 }, "currency": "USD", "observed_on": "2026-08-17", "sources": [ "gpt-5-6-luna-s2" ], "note": "Base rate. Prompts above 272K tokens are billed at 2x input and 1.5x output." } ], "undisclosed": [ "parameter_count" ] }, { "id": "gpt-5-6-terra", "model": "GPT-5.6 Terra", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "agentic", "structured_output", "function_calling" ], "tags": [], "note": "The third variant of the GPT-5.6 series.", "events": [ { "type": "announcement", "date": "2026-07-09", "sources": [ "gpt-5-6-terra-s1", "gpt-5-6-terra-s2" ] } ], "specifications": { "language": { "context_window": 1050000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-5-6-terra-s1", "url": "https://en.wikipedia.org/wiki/GPT-5.6", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20260709154055/https://en.wikipedia.org/wiki/GPT-5.6", "retrieved": "2026-07-09" }, { "id": "gpt-5-6-terra-s2", "url": "https://developers.openai.com/api/docs/models/gpt-5.6-terra", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260713173511/https://developers.openai.com/api/docs/models/gpt-5.6-terra", "retrieved": "2026-07-13" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Split from the former \"gpt-5-6\" record, which was named for the family while holding one variant's figures. Evidence: \"GPT-5.6 Terra Model — its own model page in OpenAI's API documentation, with its own specifications.\" Modalities taken from the documentation's own declaration: \"Input Text, image Output Text\". Capabilities evidenced from the primary sources: reasoning: “Reasoning token support”." }, "evidence": { "context_window": [ { "value": 1050000, "sources": [ "gpt-5-6-terra-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 2, "output": 12 }, "currency": "USD", "observed_on": "2026-08-17", "sources": [ "gpt-5-6-terra-s2" ], "note": "Base rate. Prompts above 272K tokens are billed at 2x input and 1.5x output." } ], "undisclosed": [ "parameter_count" ] }, { "id": "muse-spark-1-1", "model": "Muse Spark 1.1", "company": "Meta AI", "family": "Muse", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "reasoning" ], "tags": [], "note": "Meta Superintelligence Labs' upgrade to Muse Spark: a multimodal reasoning model built for agentic tasks, with gains in tool and computer use, coding and multimodal understanding, and a 1M-token context window. Launched with the public preview of the Meta Model API and in Thinking mode in the Meta AI app.", "events": [ { "type": "announcement", "date": "2026-07-09", "sources": [ "muse-spark-1-1-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "muse-spark-1-1-s1", "url": "https://research.meta.ai/blog/introducing-muse-spark-meta-model-api", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260805194205/https://research.meta.ai/blog/introducing-muse-spark-meta-model-api", "retrieved": "2026-08-05" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-07-09", "sources": [ "muse-spark-1-1-s1" ] } ] }, "undisclosed": [ "context_window", "parameter_count" ] }, { "id": "kimi-k3", "model": "Kimi K3", "company": "Moonshot AI", "family": "Kimi", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "coding" ], "tags": [], "note": "A 2.8-trillion-parameter open MoE, the largest open-weight model ever published.", "events": [ { "type": "announcement", "date": "2026-07-16", "sources": [ "kimi-k3-s1", "kimi-k3-s2" ] } ], "specifications": { "language": { "context_window": 1048576, "parameter_count": 2800000000000 } }, "access": { "open_weights": true, "license": "Kimi K3 License" }, "sources": [ { "id": "kimi-k3-s1", "url": "https://simonwillison.net/2026/Jul/16/kimi-k3/", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20260716202215/https://simonwillison.net/2026/Jul/16/kimi-k3/", "retrieved": "2026-07-16" }, { "id": "kimi-k3-s2", "url": "https://huggingface.co/moonshotai/Kimi-K3", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260727151224/https://huggingface.co/moonshotai/Kimi-K3", "retrieved": "2026-07-27" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the official model card. Not verified because the release date, the context window and the parameter count could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present." }, "evidence": { "release_date": [ { "value": "2026-07-16", "sources": [ "kimi-k3-s2" ] } ], "context_window": [ { "value": 1048576, "sources": [ "kimi-k3-s2" ] } ], "parameter_count": [ { "value": 2800000000000, "sources": [ "kimi-k3-s2" ] } ] } }, { "id": "qwen-image-3-0", "model": "Qwen-Image-3.0", "company": "Alibaba Qwen", "family": "Qwen-Image", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "Alibaba's third-generation image model, aimed at dense text rendering and knowledge-grounded detail.", "events": [ { "type": "announcement", "date": "2026-07-16", "sources": [ "qwen-image-3-0-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "qwen-image-3-0-s1", "url": "https://qwen.ai/blog?id=qwen-image-3.0", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260721075305/https://qwen.ai/blog?id=qwen-image-3.0", "retrieved": "2026-07-21" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present." }, "evidence": {} }, { "id": "seed-audio-1-0", "model": "Seed Audio 1.0", "company": "ByteDance", "family": "Seed Audio", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "An audio creation model covering speech through to full-scene audio generation.", "events": [ { "type": "announcement", "date": "2026-07-20", "sources": [ "seed-audio-1-0-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "seed-audio-1-0-s1", "url": "https://seed.bytedance.com/blog/from-speech-to-audio-creation-introducing-the-seed-audio-1-0-audio-creation-model", "type": "official_announcement", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." } }, { "id": "gemini-3-6-flash", "model": "Gemini 3.6 Flash", "company": "Google DeepMind", "family": "Gemini", "classification": { "primary_type": "language", "subtype": "slm" }, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "capabilities": [ "agentic", "coding", "reasoning", "function_calling", "structured_output" ], "tags": [ "small-efficient" ], "note": "Google's token-efficient workhorse update with better coding and multimodal performance.", "events": [ { "type": "announcement", "date": "2026-07-21", "sources": [ "gemini-3-6-flash-s1", "gemini-3-6-flash-s2" ] } ], "specifications": { "language": { "context_window": 1048576, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-3-6-flash-s1", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260721153557/https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/", "retrieved": "2026-07-21" }, { "id": "gemini-3-6-flash-s2", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260722113351/https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash", "retrieved": "2026-07-22" }, { "id": "gemini-3-6-flash-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260731100445/https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash?hl=en", "retrieved": "2026-07-31" }, { "id": "gemini-3-6-flash-price-20260801", "url": "https://ai.google.dev/gemini-api/docs/pricing", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260801152348/https://ai.google.dev/gemini-api/docs/pricing", "retrieved": "2026-08-01" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (Jul 21, 2026) and context window (1M) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Modalities taken from the documentation's own declaration: \"Inputs Text, Image, Video, Audio, and PDF Output Text\". Capabilities evidenced from the primary sources: agentic: “5 Flash-Lite: Built to scale agentic workflows Beyond Flash, we’re also releasing Gemini 3.”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2026-07-21", "sources": [ "gemini-3-6-flash-s1", "gemini-3-6-flash-s2" ] } ], "context_window": [ { "value": 1048576, "sources": [ "gemini-3-6-flash-s1", "gemini-3-6-flash-s2" ] } ] }, "benchmarks": [ { "name": "SWE-bench", "score": 54.2, "evaluation_type": "vendor_reported", "reported_on": "2026-07-21", "sources": [ "gemini-3-6-flash-s1" ] } ], "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 1.5, "output": 7.5 }, "currency": "USD", "observed_on": "2026-08-01", "sources": [ "gemini-3-6-flash-price-20260801" ], "note": "Paid-tier input and output as the page read on this date. The free tier and context-caching rates are priced separately and are not recorded here." } ] }, { "id": "gemini-3-5-flash-lite", "model": "Gemini 3.5 Flash-Lite", "company": "Google DeepMind", "family": "Gemini", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "function_calling", "structured_output", "agentic" ], "tags": [], "note": "The lightweight tier of the 3.5 generation, released to general availability alongside Gemini 3.6 Flash.", "events": [ { "type": "announcement", "date": "2026-07-21", "sources": [ "gemini-3-5-flash-lite-s1" ] } ], "specifications": { "language": { "context_window": 1048576, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-3-5-flash-lite-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20250605063552/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2025-06-05" }, { "id": "gemini-3-5-flash-lite-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260722151134/https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite?hl=en", "retrieved": "2026-07-22" }, { "id": "gemini-3-5-flash-lite-price-20260801", "url": "https://ai.google.dev/gemini-api/docs/pricing", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260801152348/https://ai.google.dev/gemini-api/docs/pricing", "retrieved": "2026-08-01" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date and the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 0.3, "output": 2.5 }, "currency": "USD", "observed_on": "2026-08-01", "sources": [ "gemini-3-5-flash-lite-price-20260801" ], "note": "Paid-tier input and output as the page read on this date. The free tier and context-caching rates are priced separately and are not recorded here." } ] }, { "id": "mai-image-2-5-pro", "model": "MAI-Image-2.5-Pro", "company": "Microsoft", "family": "MAI-Image", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "The higher-capability tier of MAI-Image-2.5, announced alongside MAI-Voice-2-Flash.", "events": [ { "type": "announcement", "date": "2026-07-23", "sources": [ "mai-image-2-5-pro-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mai-image-2-5-pro-s1", "url": "https://microsoft.ai/news/introducing-mai-image-2-5-pro-and-mai-voice-2-flash/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260724015709/https://microsoft.ai/news/introducing-mai-image-2-5-pro-and-mai-voice-2-flash/", "retrieved": "2026-07-24" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2026-07-23", "sources": [ "mai-image-2-5-pro-s1" ] } ] } }, { "id": "mai-voice-2-flash", "model": "MAI-Voice-2-Flash", "company": "Microsoft", "family": "MAI-Voice", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "The low-latency tier of MAI-Voice-2, announced alongside MAI-Image-2.5-Pro.", "events": [ { "type": "announcement", "date": "2026-07-23", "sources": [ "mai-voice-2-flash-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mai-voice-2-flash-s1", "url": "https://microsoft.ai/news/introducing-mai-image-2-5-pro-and-mai-voice-2-flash/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260724015709/https://microsoft.ai/news/introducing-mai-image-2-5-pro-and-mai-voice-2-flash/", "retrieved": "2026-07-24" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2026-07-23", "sources": [ "mai-voice-2-flash-s1" ] } ] } }, { "id": "flux-3", "model": "FLUX 3", "company": "Black Forest Labs", "family": "FLUX", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": null, "capabilities": [ "multilingual", "image_generation" ], "tags": [], "note": "A multimodal backbone for image, video and audio, released into early access with weights promised later.", "events": [ { "type": "announcement", "date": "2026-07-23", "sources": [ "flux-3-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "flux-3-s1", "url": "https://bfl.ai/blog/flux-3", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260723151519/https://bfl.ai/blog/flux-3", "retrieved": "2026-07-23" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there. Capabilities evidenced from the primary sources: multilingual: “b]:font-medium [&>strong]:font-medium\">While still in development, FLUX 3 Video is already particularly strong in capturing human facial expressions, associating sounds with physic” | image_generation: “(“FLUX 3 Video”) b]:font-medium [&>strong]:font-medium\">Action prediction through selected research and commercial partners, beginning with mimic robotics(“FLUX-mimic and FLUX 3 Ac”." }, "evidence": { "release_date": [ { "value": "2026-07-23", "sources": [ "flux-3-s1" ] } ] } }, { "id": "claude-opus-5", "model": "Claude Opus 5", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "agentic", "vision", "multilingual", "reasoning", "coding" ], "tags": [ "flagship" ], "note": "A step change over Opus 4.8, with a one-million-token context window and 3D rendering.", "events": [ { "type": "announcement", "date": "2026-07-24", "sources": [ "claude-opus-5-s1", "claude-opus-5-s2" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-opus-5-s1", "url": "https://www.anthropic.com/news/claude-opus-5", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260724170217/https://www.anthropic.com/news/claude-opus-5", "retrieved": "2026-07-24" }, { "id": "claude-opus-5-s2", "url": "https://platform.claude.com/docs/en/about-claude/models/overview", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260724014617/https://platform.claude.com/docs/en/about-claude/models/overview", "retrieved": "2026-07-24" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Every value this record asserts — date (Jul 24, 2026) and context window (1M) — was found in an archived primary source, checked 2026-08-10. Parameter count is not recorded, so nothing is claimed about it. Capabilities evidenced from the primary sources: agentic: “Cristian Rivera Staff Software Engineer Claude Opus 5 made large scale changes across our Fundamental Research Assistant codebase, adapting to feedback throughout an agen” | multilingual: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.” | vision: “All current Claude models support text and image input, text output, multilingual capabilities, and vision.”." }, "undisclosed": [ "parameter_count" ], "evidence": { "release_date": [ { "value": "2026-07-24", "sources": [ "claude-opus-5-s1" ] } ], "context_window": [ { "value": 1000000, "sources": [ "claude-opus-5-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 5, "output": 25 }, "currency": "USD", "sources": [ "claude-opus-5-s1" ], "observed_on": "2026-07-24" } ] }, { "id": "mai-cyber-1-flash", "model": "MAI-Cyber-1-Flash", "company": "Microsoft", "family": "MAI", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [], "tags": [], "note": "A security-focused model, announced as part of Microsoft's MDASH work.", "events": [ { "type": "announcement", "date": "2026-07-27", "sources": [ "mai-cyber-1-flash-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mai-cyber-1-flash-s1", "url": "https://microsoft.ai/news/introducing-mai-cyber-1-flash-inside-mdash/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260727172721/https://microsoft.ai/news/introducing-mai-cyber-1-flash-inside-mdash/", "retrieved": "2026-07-27" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2026-07-27", "sources": [ "mai-cyber-1-flash-s1" ] } ] }, "undisclosed": [ "context_window", "parameter_count" ] }, { "id": "minimax-h3", "model": "MiniMax H3", "company": "MiniMax", "family": "MiniMax H", "classification": { "primary_type": "video_generation", "subtype": null }, "modalities": { "input": [ "text", "image" ], "output": [ "video" ] }, "capabilities": [ "video_generation" ], "tags": [], "note": "MiniMax's video generation model, taking text and image prompts to produce video.", "events": [ { "type": "announcement", "date": "2026-07-28", "sources": [ "minimax-h3-s1", "minimax-h3-s2" ] } ], "specifications": { "language": {} }, "access": { "open_weights": true, "license": "minimax-h3-community-license-agreement" }, "sources": [ { "id": "minimax-h3-s1", "url": "https://huggingface.co/MiniMaxAI/MiniMax-H3", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260803023255/https://huggingface.co/MiniMaxAI/MiniMax-H3", "retrieved": "2026-08-03" }, { "id": "minimax-h3-s2", "url": "https://www.minimax.io/blog/minimax-h3", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260731201518/https://www.minimax.io/blog/minimax-h3", "retrieved": "2026-07-31" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the official model card and the lab's own announcement. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Licence from the model card's own license_name field on MiniMaxAI/MiniMax-H3. Capabilities evidenced from the primary sources: video_generation: “case-T2VA Type: Text-to-video Duration: 10 seconds Aspect ratio: 16:9 stage request result H3-Context-IR View script { \"task\": { \"id\": \"<task_id>\", \"model\": \"MiniMa”." }, "evidence": {} }, { "id": "grok-voice-think-fast-2-0", "model": "Grok Voice Think Fast 2.0", "company": "xAI", "family": "Grok Voice", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "xAI's most capable speech-to-speech voice model at release, served through the Speech to Speech API as grok-voice-think-fast-2.0; grok-voice-latest began routing to it on 2026-08-05.", "events": [ { "type": "announcement", "date": "2026-07-29", "sources": [ "grok-voice-think-fast-2-0-s1", "grok-voice-think-fast-2-0-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "grok-voice-think-fast-2-0-s1", "url": "https://x.ai/news/grok-voice-think-fast-2", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260730015659/https://x.ai/news/grok-voice-think-fast-2", "retrieved": "2026-07-30" }, { "id": "grok-voice-think-fast-2-0-s2", "url": "https://docs.x.ai/developers/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260725105151/https://docs.x.ai/developers/release-notes", "retrieved": "2026-07-25" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-07-29", "sources": [ "grok-voice-think-fast-2-0-s1" ] } ] } }, { "id": "gemini-robotics-er-2", "model": "Gemini Robotics-ER 2", "company": "Google DeepMind", "family": "Gemini Robotics", "classification": { "primary_type": "vision", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "An embodied reasoning endpoint for robotics, covering spatial reasoning, tool orchestration and multi-robot coordination.", "events": [ { "type": "announcement", "date": "2026-07-30", "sources": [ "gemini-robotics-er-2-s1", "gemini-robotics-er-2-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-robotics-er-2-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260729115014/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2026-07-29" }, { "id": "gemini-robotics-er-2-s2", "url": "https://deepmind.google/models/gemini-robotics/", "type": "official_documentation", "authority": "primary", "archived_url": "https://archive.md/20251120093456/https://deepmind.google/models/gemini-robotics/", "retrieved": "2025-11-20" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be traced: every archived snapshot of those sources predates the release — the newest is 2026-07-29, captured before the page announced it — so the archived copy does not contain the figure. A capture from on or after 2026-07-30 would settle it." } }, { "id": "seedance-2-5", "model": "Seedance 2.5", "company": "ByteDance", "family": "Seedance", "classification": { "primary_type": "video_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "A joint audio-video generation model built around one-take creation and flexible referencing.", "events": [ { "type": "announcement", "date": "2026-07-31", "sources": [ "seedance-2-5-s1", "seedance-2-5-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "seedance-2-5-s1", "url": "https://seed.bytedance.com/blog/one-take-creation-flexible-referencing-introducing-seedance-2-5", "type": "official_announcement", "authority": "primary", "archived_url": null, "retrieved": null }, { "id": "seedance-2-5-s2", "url": "https://seed.bytedance.com/en/seedance2_5", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260731062553/https://seed.bytedance.com/en/seedance2_5", "retrieved": "2026-07-31" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement and official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": {} }, { "id": "qwen-3-8-max", "model": "Qwen3.8-Max", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "reasoning" ], "tags": [ "flagship" ], "note": "A 2.4-trillion-parameter MoE with a 256K context, Alibaba's largest model to date. Served via API; the weights were published five days later as Qwen3.8-2.4T-A95B, tracked separately.", "events": [ { "type": "announcement", "date": "2026-08-03", "sources": [ "qwen-3-8-max-s0" ] } ], "specifications": { "language": { "context_window": 256000, "parameter_count": 2400000000000 } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "qwen-3-8-max-s0", "url": "https://qwen.ai/blog?id=qwen3.8", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260803035012/https://qwen.ai/blog?id=qwen3.8", "retrieved": "2026-08-03" }, { "id": "qwen-3-8-max-s1", "url": "https://www.marktechpost.com/2026/08/03/alibaba-qwen-releases-qwen3-8-max/", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20260803122753/https://www.marktechpost.com/2026/08/03/alibaba-qwen-releases-qwen3-8-max/", "retrieved": "2026-08-03" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Re-sourced to the lab's own announcement, replacing a news citation. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present. The promised weights were published on 2026-08-08 as Qwen3.8-2.4T-A95B and are tracked as their own record." }, "evidence": {} }, { "id": "qwen-3-8-flash-next", "model": "Qwen3.8-Flash-Next", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "Qwen introduced Qwen3.8-Flash-Next as the first open-weight release under a new architecture aimed at cost efficiency, combining Gated DeltaNet and Gated Attention with Qwen Sparse Attention and a Gated Residual stream. Qwen states a 125B-parameter main model supplemented by an additional 51B N-gram embeddings, with 6B parameters activated per token, and reports training cost around one ninth of Qwen3.7-Plus.", "events": [ { "type": "announcement", "date": "2026-08-03", "sources": [ "qwen-3-8-flash-next-s1", "qwen-3-8-flash-next-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": true, "license": "qwen-community-1.0" }, "sources": [ { "id": "qwen-3-8-flash-next-s1", "url": "https://qwen.ai/blog?id=qwen3.8-flash-next", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260826125255/https://qwen.ai/blog?id=qwen3.8-flash-next", "retrieved": "2026-08-26" }, { "id": "qwen-3-8-flash-next-s2", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260825134710/https://huggingface.co/Qwen/Qwen3.8-Flash-Next", "retrieved": "2026-08-25" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement and the official model card. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities from the model card's own pipeline_tag \"image-text-to-text\" on Qwen/Qwen3.8-Flash-Next." }, "evidence": {}, "undisclosed": [ "context_window", "parameter_count" ] }, { "id": "flux-3-video", "model": "FLUX 3 Video", "company": "Black Forest Labs", "family": "FLUX Video", "classification": { "primary_type": "video_generation", "subtype": null }, "modalities": null, "capabilities": [ "video_generation" ], "tags": [], "note": "Generation of clips up to 20 seconds at HD and Full HD, with audio created alongside the video.", "events": [ { "type": "announcement", "date": "2026-08-04", "sources": [ "flux-3-video-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "flux-3-video-s1", "url": "https://bfl.ai/blog/flux-3-video", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260805035413/https://bfl.ai/blog/flux-3-video", "retrieved": "2026-08-05" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there. Capabilities evidenced from the primary sources: video_generation: “b]:font-medium [&>strong]:font-medium\"> FLUX 3 Video provides a SOTA Video Experience b]:font-medium [&>strong]:font-medium\">FLUX 3 Video provides SOTA capabilities in both text-to”." }, "evidence": { "release_date": [ { "value": "2026-08-04", "sources": [ "flux-3-video-s1" ] } ] } }, { "id": "shieldstral-1-0-3b", "model": "Shieldstral 1.0 3B", "company": "Mistral AI", "family": "Mistral", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "A small open safety-classification model, released under Apache 2.0.", "events": [ { "type": "announcement", "date": "2026-08-04", "sources": [ "shieldstral-1-0-3b-s1", "shieldstral-1-0-3b-s2" ] }, { "type": "weights_availability", "date": "2026-08-04", "sources": [ "shieldstral-1-0-3b-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "shieldstral-1-0-3b-s1", "url": "https://mistral.ai/news/shieldstral", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260804165841/https://mistral.ai/news/shieldstral/", "retrieved": "2026-08-04" }, { "id": "shieldstral-1-0-3b-s2", "url": "https://huggingface.co/mistralai/Shieldstral-1.0-3B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260805060349/https://huggingface.co/mistralai/Shieldstral-1.0-3B", "retrieved": "2026-08-05" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement and the official model card, and every value it asserts was found there. Modalities recorded as text-to-text. The primary source states the model is text-only." }, "evidence": { "release_date": [ { "value": "2026-08-04", "sources": [ "shieldstral-1-0-3b-s1", "shieldstral-1-0-3b-s2" ] } ] }, "undisclosed": [ "parameter_count" ] }, { "id": "lfm2-5-2-6b", "model": "LFM2.5-2.6B", "company": "Liquid AI", "family": "LFM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "A 2.6B open-weight model in Liquid AI's on-device LFM2.5 line.", "events": [ { "type": "announcement", "date": "2026-08-04", "sources": [ "lfm2-5-2-6b-s1", "lfm2-5-2-6b-s2" ] }, { "type": "weights_availability", "date": "2026-08-04", "sources": [ "lfm2-5-2-6b-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 2600000000 } }, "access": { "open_weights": true, "license": "lfm1.0" }, "sources": [ { "id": "lfm2-5-2-6b-s1", "url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260805033613/https://www.liquid.ai/blog/lfm2-5-2-6b", "retrieved": "2026-08-05" }, { "id": "lfm2-5-2-6b-s2", "url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260804175207/https://huggingface.co/LiquidAI/LFM2.5-2.6B", "retrieved": "2026-08-04" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement and the official model card, and every value it asserts was found there. Modalities recorded as text-to-text. The primary source states the model is text-only. Licence from the model card's own license_name field on LiquidAI/LFM2.5-2.6B." }, "evidence": { "release_date": [ { "value": "2026-08-04", "sources": [ "lfm2-5-2-6b-s2" ] } ], "parameter_count": [ { "value": 2600000000, "sources": [ "lfm2-5-2-6b-s1", "lfm2-5-2-6b-s2" ] } ] } }, { "id": "muse-spark-1-2", "model": "Muse Spark 1.2", "company": "Meta AI", "family": "Muse", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "agentic" ], "tags": [ "flagship" ], "note": "Shipped with Muse Code, Meta's first terminal coding agent.", "events": [ { "type": "announcement", "date": "2026-08-05", "sources": [ "muse-spark-1-2-s1", "muse-spark-1-2-s2" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "muse-spark-1-2-s1", "url": "https://en.wikipedia.org/wiki/Muse_Spark", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20260718013514/https://en.wikipedia.org/wiki/Muse_Spark", "retrieved": "2026-07-18" }, { "id": "muse-spark-1-2-s2", "url": "https://artificialanalysis.ai/models/muse-spark-1-2", "type": "news", "authority": "secondary", "archived_url": "https://web.archive.org/web/20260805215022/https://artificialanalysis.ai/models/muse-spark-1-2", "retrieved": "2026-08-05" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "No primary source. The release is recorded from news reporting only, so it cannot be verified until the lab's own announcement is cited." } }, { "id": "seedrealtime", "model": "SeedRealtime", "company": "ByteDance", "family": "SeedRealtime", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "A native audio-visual full-duplex model that understands audio, visual and temporal information together.", "events": [ { "type": "announcement", "date": "2026-08-05", "sources": [ "seedrealtime-s1", "seedrealtime-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "seedrealtime-s1", "url": "https://seed.bytedance.com/blog/seedrealtime-audio-visual-full-duplex-llm-released-toward-omni-modal-natural-interaction", "type": "official_announcement", "authority": "primary", "archived_url": "https://archive.md/20260805042512/https://seed.bytedance.com/blog/seedrealtime-audio-visual-full-duplex-llm-released-toward-omni-modal-natural-interaction", "retrieved": "2026-08-05" }, { "id": "seedrealtime-s2", "url": "https://seed.bytedance.com/en/SeedRealtime", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260805042444/https://seed.bytedance.com/en/seedrealtime", "retrieved": "2026-08-05" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement and official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": { "release_date": [ { "value": "2026-08-05", "sources": [ "seedrealtime-s1" ] } ] } }, { "id": "qwen-3-8-27b", "model": "Qwen3.8-27B", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "The 27B open-weight model of the Qwen3.8 generation, released under Apache 2.0.", "events": [ { "type": "announcement", "date": "2026-08-05", "sources": [ "qwen-3-8-27b-s1" ] } ], "specifications": { "language": { "context_window": 256000, "parameter_count": 27000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "qwen-3-8-27b-s1", "url": "https://huggingface.co/Qwen/Qwen3.8-27B", "type": "official_model_card", "authority": "primary", "archived_url": "https://archive.md/20260814145858/https://huggingface.co/Qwen/Qwen3.8-27B", "retrieved": "2026-08-14" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the official model card. Not verified because the release date and the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities from the model card's own pipeline_tag \"image-text-to-text\" on Qwen/Qwen3.8-27B." }, "evidence": { "parameter_count": [ { "value": 27000000000, "sources": [ "qwen-3-8-27b-s1" ] } ] } }, { "id": "grok-imagine-image-2-0", "model": "Grok Imagine Image 2.0", "company": "xAI", "family": "Grok Imagine Image", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "xAI's image generation and editing model, served as grok-imagine-image-2.0. xAI is retiring grok-imagine-image-quality on 2026-11-02 and routing it to this model.", "events": [ { "type": "announcement", "date": "2026-08-07", "sources": [ "grok-imagine-image-2-0-s1", "grok-imagine-image-2-0-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "grok-imagine-image-2-0-s1", "url": "https://x.ai/news/grok-imagine-image-2", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260808035748/https://x.ai/news/grok-imagine-image-2", "retrieved": "2026-08-08" }, { "id": "grok-imagine-image-2-0-s2", "url": "https://docs.x.ai/developers/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260806200153/https://docs.x.ai/developers/release-notes", "retrieved": "2026-08-06" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-08-07", "sources": [ "grok-imagine-image-2-0-s1" ] } ] } }, { "id": "qwen-3-8-2-4t-a95b", "model": "Qwen3.8-2.4T-A95B", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [], "tags": [], "note": "The open weights of the Qwen3.8-Max architecture — 2.4 trillion parameters with 95B active — promised at the Max announcement and published five days later.", "events": [ { "type": "announcement", "date": "2026-08-08", "sources": [ "qwen-3-8-2-4t-a95b-s1" ] } ], "specifications": { "language": { "context_window": 256000, "parameter_count": 2400000000000 } }, "access": { "open_weights": true, "license": "qwen3.8-max" }, "sources": [ { "id": "qwen-3-8-2-4t-a95b-s1", "url": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B", "type": "official_model_card", "authority": "primary", "archived_url": "https://archive.md/20260812150420/https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B", "retrieved": "2026-08-12" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the official model card. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Licence from the model card's own license_name field on Qwen/Qwen3.8-2.4T-A95B." }, "evidence": { "context_window": [ { "value": 256000, "sources": [ "qwen-3-8-2-4t-a95b-s1" ] } ], "parameter_count": [ { "value": 2400000000000, "sources": [ "qwen-3-8-2-4t-a95b-s1" ] } ] } }, { "id": "muse-glimmer", "model": "Muse Glimmer", "company": "Meta AI", "family": "Muse", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "agentic", "tool_use", "coding", "reasoning", "multilingual", "vision" ], "tags": [], "note": "A 30B open-weights agentic model distilled from Muse Spark, sized to run on a single consumer GPU: quantised to roughly 4-bit to fit under 20GB, and shipped with a DFlash speculative-decoding drafter. Meta's pitch is the always-on local agent rather than the frontier.", "events": [ { "type": "announcement", "date": "2026-08-10", "sources": [ "muse-glimmer-s1", "muse-glimmer-s2" ] } ], "specifications": { "language": { "context_window": 131072, "parameter_count": 30000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "muse-glimmer-s1", "url": "https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260810115311/https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model", "retrieved": "2026-08-10" }, { "id": "muse-glimmer-s2", "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260810104818/https://huggingface.co/meta-models/Muse-Glimmer-30B", "retrieved": "2026-08-10" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement and the official model card. Not verified because the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities from the model card's own pipeline_tag \"image-text-to-text\" on meta-models/Muse-Glimmer-30B. Capabilities evidenced from the primary sources: agentic: “Muse Glimmer is a 30-billion-parameter model optimized for always-on local agent workflows.” | coding: “Muse Glimmer achieves strong success rates on full-task benchmarks including DeepSearch QA, MCP-Atlas, 𝛕-Bench and SWE-Bench, which measure its ability to work within sc” | reasoning: “Muse Glimmer supports different reasoning strengths to select the right balance between quality and speed.” | multilingual: “Muse Glimmer is trained on data from more than 100 languages.”." }, "evidence": { "release_date": [ { "value": "2026-08-10", "sources": [ "muse-glimmer-s1", "muse-glimmer-s2" ] } ], "context_window": [ { "value": 131072, "sources": [ "muse-glimmer-s2" ] } ], "parameter_count": [ { "value": 30000000000, "sources": [ "muse-glimmer-s1", "muse-glimmer-s2" ] } ] }, "benchmarks": [ { "name": "OSWorld-Verified", "score": 65.9, "evaluation_type": "vendor_reported", "reported_on": "2026-08-10", "sources": [ "muse-glimmer-s2" ], "compared_against": [ "Muse Glimmer", "Gemma", "Qwen3.6-27B" ] }, { "name": "SWE-bench Pro", "score": 51.2, "evaluation_type": "vendor_reported", "reported_on": "2026-08-10", "sources": [ "muse-glimmer-s2" ], "compared_against": [ "Muse Glimmer", "Gemma", "Qwen3.6-27B" ] }, { "name": "SWE-bench Verified", "score": 76, "evaluation_type": "vendor_reported", "reported_on": "2026-08-10", "sources": [ "muse-glimmer-s2" ], "compared_against": [ "Muse Glimmer", "Gemma", "Qwen3.6-27B" ] }, { "name": "SciCode", "score": 43.6, "evaluation_type": "vendor_reported", "reported_on": "2026-08-10", "sources": [ "muse-glimmer-s2" ], "compared_against": [ "Muse Glimmer", "Gemma", "Qwen3.6-27B" ] }, { "name": "ScreenSpot Pro", "score": 75.4, "evaluation_type": "vendor_reported", "reported_on": "2026-08-10", "sources": [ "muse-glimmer-s2" ], "compared_against": [ "Muse Glimmer", "Gemma", "Qwen3.6-27B" ] }, { "name": "IFBench", "score": 77, "evaluation_type": "vendor_reported", "reported_on": "2026-08-10", "sources": [ "muse-glimmer-s2" ], "compared_against": [ "Muse Glimmer", "Gemma", "Qwen3.6-27B" ] }, { "name": "AIME", "score": 94.7, "evaluation_type": "vendor_reported", "reported_on": "2026-08-10", "sources": [ "muse-glimmer-s2" ], "compared_against": [ "Muse Glimmer", "Gemma", "Qwen3.6-27B" ] } ] }, { "id": "mai-image-2-6", "model": "MAI-Image-2.6", "company": "Microsoft", "family": "MAI-Image", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "The current MAI image generation and editing model.", "events": [ { "type": "announcement", "date": "2026-08-10", "sources": [ "mai-image-2-6-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mai-image-2-6-s1", "url": "https://microsoft.ai/news/mai-image-2-6-launches-at-no-2-on-arena-ahead-of-google-meta-and-xai/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260813024921/https://microsoft.ai/news/mai-image-2-6-launches-at-no-2-on-arena-ahead-of-google-meta-and-xai/", "retrieved": "2026-08-13" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2026-08-10", "sources": [ "mai-image-2-6-s1" ] } ] } }, { "id": "gpt-5-6-cyber", "model": "GPT-5.6-Cyber", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "structured_output", "function_calling", "reasoning" ], "tags": [], "note": "A cybersecurity-specific model available only through Daybreak Red, OpenAI's gated access tier for approved defenders.", "events": [ { "type": "announcement", "date": "2026-08-10", "sources": [ "gpt-5-6-cyber-s1" ] } ], "specifications": { "language": { "context_window": 400000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-5-6-cyber-s1", "url": "https://openai.com/index/expanding-daybreak-as-the-cyber-defense-window-narrows/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260810220432/https://openai.com/index/expanding-daybreak-as-the-cyber-defense-window-narrows/", "retrieved": "2026-08-10" }, { "id": "gpt-5-6-cyber-odocs", "url": "https://developers.openai.com/api/docs/models/gpt-5.6-cyber", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260812140241/https://developers.openai.com/api/docs/models/gpt-5.6-cyber", "retrieved": "2026-08-12" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2026-08-10", "sources": [ "gpt-5-6-cyber-s1" ] } ], "context_window": [ { "value": 400000, "sources": [ "gpt-5-6-cyber-odocs" ] } ] }, "undisclosed": [ "parameter_count" ], "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 12.5, "output": 75 }, "currency": "USD", "observed_on": "2026-10-15", "sources": [ "gpt-5-6-cyber-odocs" ], "note": "Base rate. Prompts above 272K tokens are billed at 2x input and 1.5x output." } ] }, { "id": "nemotron-3-5-lightning", "model": "Nemotron 3.5 Lightning", "company": "NVIDIA", "family": "Nemotron", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "agentic" ], "tags": [], "note": "A 30B mixture-of-experts model with 3B active parameters, released with permissive licensing, weights, data and recipes, and aimed at long-running agent harnesses.", "events": [ { "type": "announcement", "date": "2026-08-11", "sources": [ "nemotron-3-5-lightning-s1" ] }, { "type": "weights_availability", "date": "2026-08-11", "sources": [ "nemotron-3-5-lightning-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 30000000000 } }, "access": { "open_weights": true, "license": null }, "sources": [ { "id": "nemotron-3-5-lightning-s1", "url": "https://developer.nvidia.com/blog/nvidia-nemotron-3-5-lightning-delivers-fast-accurate-specialized-task-execution-for-long-running-agents/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260812102612/https://developer.nvidia.com/blog/nvidia-nemotron-3-5-lightning-delivers-fast-accurate-specialized-task-execution-for-long-running-agents/", "retrieved": "2026-08-12" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2026-08-11", "sources": [ "nemotron-3-5-lightning-s1" ] } ], "parameter_count": [ { "value": 30000000000, "sources": [ "nemotron-3-5-lightning-s1" ] } ] }, "undisclosed": [ "context_window", "license" ] }, { "id": "mai-code-1-1-flash", "model": "MAI-Code-1.1-Flash", "company": "Microsoft", "family": "MAI", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "coding" ], "tags": [], "note": "An iteration of MAI-Code-1-Flash, positioned as faster at a quarter of the cost.", "events": [ { "type": "announcement", "date": "2026-08-11", "sources": [ "mai-code-1-1-flash-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mai-code-1-1-flash-s1", "url": "https://microsoft.ai/news/mai-code-1-1-flash-br-better-faster-at-a-quarter-of-the-cost/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260811194735/https://microsoft.ai/news/mai-code-1-1-flash-br-better-faster-at-a-quarter-of-the-cost/", "retrieved": "2026-08-11" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2026-08-11", "sources": [ "mai-code-1-1-flash-s1" ] } ] }, "undisclosed": [ "context_window", "parameter_count" ] }, { "id": "grok-4-6", "model": "Grok 4.6", "company": "xAI", "family": "Grok", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "agentic", "function_calling", "structured_output", "reasoning", "coding" ], "tags": [], "note": "Builds on Grok 4.5 with a focus on long-running agents and more ambitious interactive and visual work.", "events": [ { "type": "announcement", "date": "2026-08-12", "sources": [ "grok-4-6-s1", "grok-4-6-s2" ] } ], "specifications": { "language": { "context_window": 500000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "grok-4-6-s1", "url": "https://x.ai/news/grok-4-6", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260812154457/https://x.ai/news/grok-4-6", "retrieved": "2026-08-12" }, { "id": "grok-4-6-s2", "url": "https://docs.x.ai/developers/grok-4-6", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260814040702/https://docs.x.ai/developers/grok-4-6", "retrieved": "2026-08-14" }, { "id": "grok-4-6-xmodel", "url": "https://docs.x.ai/developers/models/grok-4.6", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260812154738/https://docs.x.ai/developers/models/grok-4.6", "retrieved": "2026-08-12" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement and official documentation, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2026-08-12", "sources": [ "grok-4-6-s1", "grok-4-6-s2" ] } ], "context_window": [ { "value": 500000, "sources": [ "grok-4-6-s2" ] } ] }, "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 2, "output": 6 }, "currency": "USD", "observed_on": "2026-08-17", "sources": [ "grok-4-6-s1" ], "note": "Base tier. xAI bills a higher rate for prompts at or above 200k tokens." } ], "undisclosed": [ "parameter_count" ] }, { "id": "mai-thinking-1", "model": "MAI-Thinking-1", "company": "Microsoft", "family": "MAI", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "reasoning", "coding", "function_calling" ], "tags": [], "note": "Microsoft Superintelligence team's first reasoning model: a sparse Mixture of Experts with 35B active parameters of roughly 1T total, a 256k-token window, and function calling.", "events": [ { "type": "announcement", "date": "2026-08-12", "sources": [ "mai-thinking-1-s1" ] } ], "specifications": { "language": { "context_window": 256000, "parameter_count": 35000000000 } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mai-thinking-1-s1", "url": "https://microsoft.ai/news/introducing-mai-thinking-1/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260805034757/https://microsoft.ai/news/introducing-mai-thinking-1/", "retrieved": "2026-08-05" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement. the release date could not be traced yet: this release is 1 day old and archive.org has not captured the announcement since it went up. That is a wait, not a gap — re-running the archive pass will settle it." }, "evidence": { "context_window": [ { "value": 256000, "sources": [ "mai-thinking-1-s1" ] } ], "parameter_count": [ { "value": 35000000000, "sources": [ "mai-thinking-1-s1" ] } ] }, "benchmarks": [ { "name": "AIME", "score": 97, "evaluation_type": "vendor_reported", "reported_on": "2026-08-12", "sources": [ "mai-thinking-1-s1" ] } ] }, { "id": "gemini-3-7-flash", "model": "Gemini 3.7 Flash", "company": "Google DeepMind", "family": "Gemini", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "function_calling", "structured_output", "agentic" ], "tags": [], "note": "Google's workhorse model, introduced at half the per-token price of 3.6 Flash and aimed at software engineering and agent workloads.", "events": [ { "type": "announcement", "date": "2026-08-13", "sources": [ "gemini-3-7-flash-s1" ] } ], "specifications": { "language": { "context_window": 1048576, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-3-7-flash-s1", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260813195326/https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash/", "retrieved": "2026-08-13" }, { "id": "gemini-3-7-flash-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.7-flash?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260814155256/https://ai.google.dev/gemini-api/docs/models/gemini-3.7-flash?hl=en", "retrieved": "2026-08-14" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2026-08-13", "sources": [ "gemini-3-7-flash-s1" ] } ], "context_window": [ { "value": 1048576, "sources": [ "gemini-3-7-flash-s1" ] } ] }, "undisclosed": [ "parameter_count" ] }, { "id": "glm-5-3", "model": "GLM-5.3", "company": "Zhipu AI", "family": "GLM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "coding", "agentic", "reasoning" ], "tags": [], "note": "Zhipu calls it the most capable open-weights coding model, but says three times that the weights follow two weeks after launch, once safety evaluation and hardening complete. Recorded proprietary until they land.", "events": [ { "type": "announcement", "date": "2026-08-14", "sources": [ "glm-5-3-s1", "glm-5-3-s2" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "glm-5-3-s1", "url": "https://z.ai/blog/glm-5.3", "type": "official_announcement", "authority": "primary", "archived_url": "https://archive.md/20260814053329/https://z.ai/blog/glm-5.3", "retrieved": "2026-08-14" }, { "id": "glm-5-3-s2", "url": "https://docs.z.ai/guides/llm/glm-5.3", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260814120744/https://docs.z.ai/guides/llm/glm-5.3", "retrieved": "2026-08-14" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-08-14", "sources": [ "glm-5-3-s1" ] } ], "context_window": [ { "value": 1000000, "sources": [ "glm-5-3-s1", "glm-5-3-s2" ] } ] } }, { "id": "ornith-1-5-9b", "model": "Ornith-1.5-9B", "company": "Ornith", "family": "Ornith", "classification": { "primary_type": "language", "subtype": "reasoning" }, "modalities": null, "capabilities": [ "coding", "agentic", "reasoning" ], "tags": [ "small-efficient" ], "note": "The compact dense member of Ornith-1.5, which extends the self-scaffolding loop of 1.0 into end-to-end self-improvement over generated tasks.", "events": [ { "type": "announcement", "date": "2026-08-19", "sources": [ "ornith-1-5-9b-s1", "ornith-1-5-9b-s2", "ornith-1-5-9b-s3" ] } ], "specifications": { "language": { "context_window": 262144, "parameter_count": 9000000000 } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "ornith-1-5-9b-s1", "url": "https://ornith.ai/ornith_1_5.html", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260819162854/https://ornith.ai/ornith_1_5.html", "retrieved": "2026-08-19" }, { "id": "ornith-1-5-9b-s2", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-9B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260819162709/https://huggingface.co/ornith-ai/Ornith-1.5-9B", "retrieved": "2026-08-19" }, { "id": "ornith-1-5-9b-s3", "url": "https://x.com/ornith_/status/2090074077084127302", "type": "official_announcement", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-08-19", "sources": [ "ornith-1-5-9b-s2" ] } ], "context_window": [ { "value": 262144, "sources": [ "ornith-1-5-9b-s1", "ornith-1-5-9b-s2" ] } ], "parameter_count": [ { "value": 9000000000, "sources": [ "ornith-1-5-9b-s1", "ornith-1-5-9b-s2" ] } ] } }, { "id": "ornith-1-5-35b-a3b", "model": "Ornith-1.5-35B-A3B", "company": "Ornith", "family": "Ornith", "classification": { "primary_type": "language", "subtype": "reasoning" }, "modalities": null, "capabilities": [ "coding", "agentic", "reasoning" ], "tags": [], "note": "The mid-size mixture-of-experts member of Ornith-1.5, activating about 3B parameters per token.", "events": [ { "type": "announcement", "date": "2026-08-19", "sources": [ "ornith-1-5-35b-a3b-s1", "ornith-1-5-35b-a3b-s2", "ornith-1-5-35b-a3b-s3" ] } ], "specifications": { "language": { "context_window": 262144, "parameter_count": 35000000000 } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "ornith-1-5-35b-a3b-s1", "url": "https://ornith.ai/ornith_1_5.html", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260819162854/https://ornith.ai/ornith_1_5.html", "retrieved": "2026-08-19" }, { "id": "ornith-1-5-35b-a3b-s2", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260819141646/https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B", "retrieved": "2026-08-19" }, { "id": "ornith-1-5-35b-a3b-s3", "url": "https://x.com/ornith_/status/2090074077084127302", "type": "official_announcement", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-08-19", "sources": [ "ornith-1-5-35b-a3b-s2" ] } ], "context_window": [ { "value": 262144, "sources": [ "ornith-1-5-35b-a3b-s1", "ornith-1-5-35b-a3b-s2" ] } ], "parameter_count": [ { "value": 35000000000, "sources": [ "ornith-1-5-35b-a3b-s1", "ornith-1-5-35b-a3b-s2" ] } ] } }, { "id": "ornith-1-5-397b", "model": "Ornith-1.5-397B", "company": "Ornith", "family": "Ornith", "classification": { "primary_type": "language", "subtype": "reasoning" }, "modalities": null, "capabilities": [ "coding", "agentic", "reasoning" ], "tags": [ "flagship" ], "note": "The frontier-scale mixture-of-experts member of Ornith-1.5, which the lab reports as on par with Claude Opus 4.8 on Terminal-Bench 2.1 and DeepSWE.", "events": [ { "type": "announcement", "date": "2026-08-19", "sources": [ "ornith-1-5-397b-s1", "ornith-1-5-397b-s2", "ornith-1-5-397b-s3" ] } ], "specifications": { "language": { "context_window": 262144, "parameter_count": 397000000000 } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "ornith-1-5-397b-s1", "url": "https://ornith.ai/ornith_1_5.html", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260819162854/https://ornith.ai/ornith_1_5.html", "retrieved": "2026-08-19" }, { "id": "ornith-1-5-397b-s2", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-397B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260820050925/https://huggingface.co/ornith-ai/Ornith-1.5-397B", "retrieved": "2026-08-20" }, { "id": "ornith-1-5-397b-s3", "url": "https://x.com/ornith_/status/2090074077084127302", "type": "official_announcement", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement and the official model card. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": { "context_window": [ { "value": 262144, "sources": [ "ornith-1-5-397b-s1", "ornith-1-5-397b-s2" ] } ], "parameter_count": [ { "value": 397000000000, "sources": [ "ornith-1-5-397b-s1", "ornith-1-5-397b-s2" ] } ] } }, { "id": "deepseek-v4-flash-vision-exp", "model": "DeepSeek-V4-Flash-Vision-Exp", "company": "DeepSeek", "family": "DeepSeek Vision", "classification": { "primary_type": "vision", "subtype": null }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "DeepSeek released DeepSeek-V4-Flash-Vision-Exp as an experimental multimodal vision understanding model on the DeepSeek API platform. It was retired on 2026-09-10: the model name is still accepted, but requests are served by DeepSeek-V4.1-Flash and billed at the Flash price.", "events": [ { "type": "announcement", "date": "2026-08-21", "sources": [ "deepseek-v4-flash-vision-exp-s1", "deepseek-v4-flash-vision-exp-s2" ] }, { "type": "retirement", "date": "2026-09-10", "sources": [ "deepseek-v4-flash-vision-exp-s1" ] } ], "specifications": {}, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "deepseek-v4-flash-vision-exp-s1", "url": "https://api-docs.deepseek.com/updates", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260821045516/https://api-docs.deepseek.com/updates/", "retrieved": "2026-08-21" }, { "id": "deepseek-v4-flash-vision-exp-s2", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260831103428/https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp", "retrieved": "2026-08-31" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation and the official model card. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities from the model card's own pipeline_tag \"image-text-to-text\" on deepseek-ai/DeepSeek-V4-Flash-Vision-Exp." }, "evidence": {} }, { "id": "glm-5-3-flash", "model": "GLM-5.3-Flash", "company": "Zhipu AI", "family": "GLM", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "Z.AI released GLM-5.3-Flash with native visual capabilities for observing interfaces, rendering results and interaction feedback, and a hybrid architecture combining linear and sparse attention with 320B total parameters and 18B activated.", "events": [ { "type": "announcement", "date": "2026-08-26", "sources": [ "glm-5-3-flash-s1", "glm-5-3-flash-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 320000000000 } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "glm-5-3-flash-s1", "url": "https://docs.z.ai/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null }, { "id": "glm-5-3-flash-s2", "url": "https://huggingface.co/zai-org/GLM-5.3-Flash", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260826140926/https://huggingface.co/zai-org/GLM-5.3-Flash", "retrieved": "2026-08-26" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`. Modalities from the model card's own pipeline_tag \"image-text-to-text\" on zai-org/GLM-5.3-Flash." }, "evidence": { "release_date": [ { "value": "2026-08-26", "sources": [ "glm-5-3-flash-s2" ] } ], "parameter_count": [ { "value": 320000000000, "sources": [ "glm-5-3-flash-s2" ] } ] } }, { "id": "gemini-3-5-transcribe", "model": "Gemini 3.5 Transcribe", "company": "Google DeepMind", "family": "Gemini Transcribe", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "Google made Gemini 3.5 Transcribe (gemini-3.5-transcribe) generally available: a dedicated non-streaming speech-to-text model based on Gemini's audio understanding, with utterance-based language detection across 85+ languages, speaker diarization, word-level timestamps and custom vocabulary biasing up to 1,000 terms.", "events": [ { "type": "announcement", "date": "2026-08-26", "sources": [ "gemini-3-5-transcribe-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-3-5-transcribe-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260826034017/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2026-08-26" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": {} }, { "id": "gemini-3-5-transcribe-live", "model": "Gemini 3.5 Transcribe Live", "company": "Google DeepMind", "family": "Gemini Transcribe", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "Released with Gemini 3.5 Transcribe: a low-latency, bidirectional streaming speech-to-text model (gemini-3.5-transcribe-live).", "events": [ { "type": "announcement", "date": "2026-08-26", "sources": [ "gemini-3-5-transcribe-live-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-3-5-transcribe-live-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260826034017/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2026-08-26" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": {} }, { "id": "hy4-preview", "model": "Hy4 preview", "company": "Tencent", "family": "Hy", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "coding", "agentic" ], "tags": [], "note": "Tencent released and open-sourced Hy4 preview for coding, office work and scientific research: 770 billion total parameters, 49 billion active, and a context window Tencent describes as exceeding 1M tokens (1,048,576 in the published config). Available on WorkBuddy, CodeBuddy, Yuanbao and ima, and by API through Tencent Cloud TokenHub and OpenRouter.", "events": [ { "type": "announcement", "date": "2026-08-28", "sources": [ "hy4-preview-s1", "hy4-preview-s2" ] } ], "specifications": { "language": { "context_window": 1048576, "parameter_count": 770000000000 } }, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "hy4-preview-s1", "url": "https://www.tencent.com/tencent-releases-and-open-sources-tencent-hy4-preview/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260828123217/https://www.tencent.com/tencent-releases-and-open-sources-tencent-hy4-preview/", "retrieved": "2026-08-28" }, { "id": "hy4-preview-s2", "url": "https://huggingface.co/tencent/Hy4-preview", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260828064930/https://huggingface.co/tencent/Hy4-preview", "retrieved": "2026-08-28" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-08-28", "sources": [ "hy4-preview-s2" ] } ], "context_window": [ { "value": 1048576, "sources": [ "hy4-preview-s2" ] } ], "parameter_count": [ { "value": 770000000000, "sources": [ "hy4-preview-s2" ] } ] } }, { "id": "claude-fable-5-1", "model": "Claude Fable 5.1", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "agentic" ], "tags": [], "note": "Anthropic launched Claude Fable 5.1 as the successor to Claude Fable 5 for long-running agentic coding, knowledge work and research. Its API id is claude-fable-5-1, and Anthropic states a 1M token context window by default with 128k maximum output tokens.", "events": [ { "type": "announcement", "date": "2026-09-01", "sources": [ "claude-fable-5-1-s1", "claude-fable-5-1-s2" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-fable-5-1-s1", "url": "https://platform.claude.com/docs/en/release-notes/api", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null }, { "id": "claude-fable-5-1-s2", "url": "https://platform.claude.com/docs/en/about-claude/models", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null }, { "id": "claude-fable-5-1-s3", "url": "https://www.anthropic.com/claude-fable-and-mythos-5-1", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260901180235/https://www.anthropic.com/claude-fable-and-mythos-5-1", "retrieved": "2026-09-01" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation and the lab's own announcement. Not verified because the release date and the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": {} }, { "id": "claude-mythos-5-1", "model": "Claude Mythos 5.1", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [], "tags": [], "note": "Anthropic launched Claude Mythos 5.1 alongside Claude Fable 5.1, with API id claude-mythos-5-1, for Project Glasswing participants. Anthropic states a 1M token context window by default with 128k maximum output tokens.", "events": [ { "type": "announcement", "date": "2026-09-01", "sources": [ "claude-mythos-5-1-s1" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-mythos-5-1-s1", "url": "https://platform.claude.com/docs/en/release-notes/api", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null }, { "id": "claude-mythos-5-1-s2", "url": "https://www.anthropic.com/claude-fable-and-mythos-5-1", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260901180235/https://www.anthropic.com/claude-fable-and-mythos-5-1", "retrieved": "2026-09-01" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation and the lab's own announcement. Not verified because the release date and the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": {} }, { "id": "muse-voice-transcribe", "model": "Muse Voice Transcribe", "company": "Meta AI", "family": "Muse Voice", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "Meta introduced Muse Voice Transcribe, a speech-to-text model with a live transcription demo on its research site.", "events": [ { "type": "announcement", "date": "2026-09-01", "sources": [ "muse-voice-transcribe-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "muse-voice-transcribe-s1", "url": "https://research.meta.ai/blog/introducing-muse-voice-transcribe", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260901181222/https://research.meta.ai/blog/introducing-muse-voice-transcribe", "retrieved": "2026-09-01" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-09-01", "sources": [ "muse-voice-transcribe-s1" ] } ] } }, { "id": "gemini-3-8-flash", "model": "Gemini 3.8 Flash", "company": "Google DeepMind", "family": "Gemini", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "function_calling", "structured_output", "agentic" ], "tags": [], "note": "Google made Gemini 3.8 Flash generally available, describing it as its most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents and complex enterprise workflows. Its API endpoint is gemini-3.8-flash.", "events": [ { "type": "announcement", "date": "2026-09-02", "sources": [ "gemini-3-8-flash-s1", "gemini-3-8-flash-s2" ] } ], "specifications": { "language": { "context_window": 1048576, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-3-8-flash-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260826034017/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2026-08-26" }, { "id": "gemini-3-8-flash-s2", "url": "https://ai.google.dev/gemini-api/docs/models", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260902150821/https://ai.google.dev/gemini-api/docs/models", "retrieved": "2026-09-02" }, { "id": "gemini-3-8-flash-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.8-flash?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "partially_verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`. Downgraded: context_window comes from the lab's documentation, but attribute-facts could not read an archived snapshot that states it, so the figure is not yet traced to a quoted sentence. reconcile-status upgrades it once attribute-facts succeeds." }, "evidence": { "release_date": [ { "value": "2026-09-02", "sources": [ "gemini-3-8-flash-s2" ] } ] } }, { "id": "muse-spark-1-3", "model": "Muse Spark 1.3", "company": "Meta AI", "family": "Muse", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [], "tags": [], "note": "Meta released Muse Spark 1.3 in Muse Code and the Meta Model API, reporting improved performance on agentic and coding tasks and, in its own engineering evaluation, ~20% fewer tool calls and ~25% fewer tokens. Previously available reasoning modes shipped on the day, with max reasoning to follow after further safety testing. Meta lists a Muse Spark open-weights release as future roadmap, so this release is not open weights.", "events": [ { "type": "announcement", "date": "2026-09-02", "sources": [ "muse-spark-1-3-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "muse-spark-1-3-s1", "url": "https://research.meta.ai/blog/introducing-muse-spark-1-3", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260902193756/https://research.meta.ai/blog/introducing-muse-spark-1-3", "retrieved": "2026-09-02" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-09-02", "sources": [ "muse-spark-1-3-s1" ] } ] } }, { "id": "gemini-3-8-flash-cyber", "model": "Gemini 3.8 Flash Cyber", "company": "Google DeepMind", "family": "Gemini", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "function_calling", "structured_output", "agentic" ], "tags": [], "note": "Google launched Gemini 3.8 Flash Cyber, which it calls its most advanced cyber model, through the Fairwind Program — a limited-access programme for governments and trusted partners — paired with its CodeMender harness to find, verify and fix vulnerabilities. It is not generally available and does not appear in the public Gemini API model list.", "events": [ { "type": "announcement", "date": "2026-09-02", "sources": [ "gemini-3-8-flash-cyber-s1" ] } ], "specifications": { "language": { "context_window": 1048576, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-3-8-flash-cyber-s1", "url": "https://blog.google/innovation-and-ai/technology/safety-security/fairwind-program/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260902180846/https://blog.google/innovation-and-ai/technology/safety-security/fairwind-program/", "retrieved": "2026-09-02" }, { "id": "gemini-3-8-flash-cyber-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.8-flash?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "partially_verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`. Downgraded: context_window comes from the lab's documentation, but attribute-facts could not read an archived snapshot that states it, so the figure is not yet traced to a quoted sentence. reconcile-status upgrades it once attribute-facts succeeds." }, "evidence": { "release_date": [ { "value": "2026-09-02", "sources": [ "gemini-3-8-flash-cyber-s1" ] } ] } }, { "id": "gpt-6-astra", "model": "GPT-6 Astra", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "structured_output", "function_calling", "reasoning" ], "tags": [], "note": "OpenAI introduced GPT-6 Astra, which it describes as its most intelligent and aligned model and state of the art on computer use, browsing, software engineering, cybersecurity, science and professional work. It rolled out first to a limited set of organizations, with ChatGPT Plus, Pro, Business and Enterprise, the API, Azure and Bedrock to follow. OpenAI's own announcement page carries no date; the date here is the one OpenAI's news index prints against the accompanying safety overview and system card.", "events": [ { "type": "announcement", "date": "2026-09-03", "sources": [ "gpt-6-astra-s1", "gpt-6-astra-s2", "gpt-6-astra-s3" ] } ], "specifications": { "language": { "context_window": 1050000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-6-astra-s1", "url": "https://openai.com/index/gpt-6-astra/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260903182810/https://openai.com/index/gpt-6-astra/", "retrieved": "2026-09-03" }, { "id": "gpt-6-astra-s2", "url": "https://openai.com/index/safety-overview-gpt-6-astra/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260904150411/https://openai.com/index/safety-overview-gpt-6-astra/", "retrieved": "2026-09-04" }, { "id": "gpt-6-astra-s3", "url": "https://deploymentsafety.openai.com/gpt-6-astra", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260903193434/https://deploymentsafety.openai.com/gpt-6-astra", "retrieved": "2026-09-03" }, { "id": "gpt-6-astra-odocs", "url": "https://developers.openai.com/api/docs/models/gpt-6-astra", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement and official documentation. Not verified because the release date and the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." } }, { "id": "lyria-3-5", "model": "Lyria 3.5", "company": "Google DeepMind", "family": "Lyria", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "Google made Lyria 3.5 (lyria-3.5) generally available as the next generation of its music generation model: full-length songs with improved musical coherence, natural vocals, and control over duration and structure.", "events": [ { "type": "announcement", "date": "2026-09-03", "sources": [ "lyria-3-5-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "lyria-3-5-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260904192757/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2026-09-04" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-09-03", "sources": [ "lyria-3-5-s1" ] } ] } }, { "id": "mai-transcribe-2", "model": "MAI-Transcribe-2", "company": "Microsoft", "family": "MAI-Transcribe", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "Microsoft AI's second-generation speech recognition model, transcribing 60 languages with speaker diarization, word-level timestamps and keyword biasing. Available in Microsoft Foundry, the MAI Playground and OpenRouter.", "events": [ { "type": "announcement", "date": "2026-09-03", "sources": [ "mai-transcribe-2-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mai-transcribe-2-s1", "url": "https://microsoft.ai/news/mai-transcribe-2-is-the-fastest-most-accurate-and-cheapest-speech-recognition-model-in-the-world/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260903162812/https://microsoft.ai/news/mai-transcribe-2-is-the-fastest-most-accurate-and-cheapest-speech-recognition-model-in-the-world/", "retrieved": "2026-09-03" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-09-03", "sources": [ "mai-transcribe-2-s1" ] } ] } }, { "id": "chatgpt-images-2-5", "model": "ChatGPT Images 2.5", "company": "OpenAI", "family": "GPT-Image", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "OpenAI introduced ChatGPT Images 2.5, which it calls a new state-of-the-art image model with sharper details, more precise editing and faster generation. OpenAI states image generation latency is reduced by up to 50% compared with Images 2.0. Available to all ChatGPT, ChatGPT Work and Codex users across desktop, mobile and web.", "events": [ { "type": "announcement", "date": "2026-09-08", "sources": [ "chatgpt-images-2-5-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "chatgpt-images-2-5-s1", "url": "https://openai.com/index/introducing-chatgpt-images-2-5/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260908183908/https://openai.com/index/introducing-chatgpt-images-2-5/", "retrieved": "2026-09-08" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-09-08", "sources": [ "chatgpt-images-2-5-s1" ] } ] } }, { "id": "gpt-image-2-5-flare", "model": "GPT-Image-2.5 Flare", "company": "OpenAI", "family": "GPT-Image", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": { "input": [ "text", "image" ], "output": [ "image" ] }, "capabilities": [], "tags": [], "note": "One of two API models OpenAI introduced alongside ChatGPT Images 2.5. OpenAI states Flare brings the same improvements in quality, editing and speed.", "events": [ { "type": "announcement", "date": "2026-09-08", "sources": [ "gpt-image-2-5-flare-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-image-2-5-flare-s1", "url": "https://openai.com/index/introducing-chatgpt-images-2-5/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260908183908/https://openai.com/index/introducing-chatgpt-images-2-5/", "retrieved": "2026-09-08" }, { "id": "gpt-image-2-5-flare-odocs", "url": "https://developers.openai.com/api/docs/models/gpt-image-2.5-flare", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-09-08", "sources": [ "gpt-image-2-5-flare-s1" ] } ] } }, { "id": "gpt-image-2-5-sunburst", "model": "GPT-Image-2.5 Sunburst", "company": "OpenAI", "family": "GPT-Image", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": { "input": [ "text", "image" ], "output": [ "image" ] }, "capabilities": [], "tags": [], "note": "The second API model introduced alongside ChatGPT Images 2.5. OpenAI states Sunburst offers an extra level of precision for detailed creative work, with longer generation times.", "events": [ { "type": "announcement", "date": "2026-09-08", "sources": [ "gpt-image-2-5-sunburst-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-image-2-5-sunburst-s1", "url": "https://openai.com/index/introducing-chatgpt-images-2-5/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260908183908/https://openai.com/index/introducing-chatgpt-images-2-5/", "retrieved": "2026-09-08" }, { "id": "gpt-image-2-5-sunburst-odocs", "url": "https://developers.openai.com/api/docs/models/gpt-image-2.5-sunburst", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-09-08", "sources": [ "gpt-image-2-5-sunburst-s1" ] } ] } }, { "id": "gpt-rosalind", "model": "GPT-Rosalind", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "OpenAI made GPT-Rosalind (gpt-rosalind-research) generally available through its trusted-access programme, for approved internal life sciences research. The date is general availability as OpenAI's API changelog states it; any earlier limited preview is not recorded here.", "events": [ { "type": "announcement", "date": "2026-09-08", "sources": [ "gpt-rosalind-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-rosalind-s1", "url": "https://developers.openai.com/api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260908093131/https://developers.openai.com/api/docs/changelog", "retrieved": "2026-09-08" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. The primary source states the model is text-only." }, "evidence": {}, "undisclosed": [ "parameter_count" ] }, { "id": "auk", "model": "AuK", "company": "Tencent", "family": "AuK", "classification": { "primary_type": "audio", "subtype": null }, "modalities": { "input": [ "text" ], "output": [ "audio" ] }, "capabilities": [ "speech_generation" ], "tags": [], "note": "Tencent open-sourced AuK, a text-to-speech model, with code and weights publicly available. Its Hugging Face repository was created on 2026-08-18, three weeks before the release.", "events": [ { "type": "announcement", "date": "2026-09-09", "sources": [ "auk-s1", "auk-s2" ] } ], "specifications": {}, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "auk-s1", "url": "https://github.com/Tencent-Hunyuan/AuK", "type": "official_repository", "authority": "primary", "archived_url": "https://web.archive.org/web/20260913062749/https://github.com/Tencent-Hunyuan/AuK", "retrieved": "2026-09-13" }, { "id": "auk-s2", "url": "https://huggingface.co/tencent/AuK", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260909040001/https://huggingface.co/tencent/AuK", "retrieved": "2026-09-09" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`. Modalities from the model card's own pipeline_tag \"text-to-speech\" on tencent/AuK. Capabilities evidenced from the primary sources: speech_generation: “toml View all files Repository files navigation README Code of conduct Contributing License More items AuK: An Open-Source Foundational Model for Speech Generation and Editing 💻 T”." }, "evidence": { "release_date": [ { "value": "2026-09-09", "sources": [ "auk-s1" ] } ] } }, { "id": "auk-flash", "model": "AuK-Flash", "company": "Tencent", "family": "AuK", "classification": { "primary_type": "audio", "subtype": null }, "modalities": { "input": [ "text" ], "output": [ "audio" ] }, "capabilities": [], "tags": [], "note": "Released alongside AuK: Tencent's README records day-zero inference support for both AuK and AuK-Flash on 2026-09-09.", "events": [ { "type": "announcement", "date": "2026-09-09", "sources": [ "auk-flash-s1", "auk-flash-s2" ] } ], "specifications": {}, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "auk-flash-s1", "url": "https://github.com/Tencent-Hunyuan/AuK", "type": "official_repository", "authority": "primary", "archived_url": "https://web.archive.org/web/20260913062749/https://github.com/Tencent-Hunyuan/AuK", "retrieved": "2026-09-13" }, { "id": "auk-flash-s2", "url": "https://huggingface.co/tencent/AuK-Flash", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260909025955/https://huggingface.co/tencent/AuK-Flash", "retrieved": "2026-09-09" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`. Modalities from the model card's own pipeline_tag \"text-to-speech\" on tencent/AuK-Flash." }, "evidence": { "release_date": [ { "value": "2026-09-09", "sources": [ "auk-flash-s1" ] } ] } }, { "id": "deepseek-v4-1-flash", "model": "DeepSeek-V4.1-Flash", "company": "DeepSeek", "family": "DeepSeek", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "DeepSeek released DeepSeek-V4.1-Flash, which it describes as the smallest model in a new architecture family, with native multimodal visual understanding and an architecture designed for a higher capability ceiling, faster inference and higher throughput. It is served on the DeepSeek API as deepseek-flash. DeepSeek also retired deepseek-v4-flash-vision-exp on the same day, with its requests now served by this model.", "events": [ { "type": "announcement", "date": "2026-09-10", "sources": [ "deepseek-v4-1-flash-s1", "deepseek-v4-1-flash-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "deepseek-v4-1-flash-s1", "url": "https://api-docs.deepseek.com/updates", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260910074600/https://api-docs.deepseek.com/updates/", "retrieved": "2026-09-10" }, { "id": "deepseek-v4-1-flash-s2", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260910055823/https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash", "retrieved": "2026-09-10" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`. Modalities from the model card's own pipeline_tag \"image-text-to-text\" on deepseek-ai/DeepSeek-V4.1-Flash." }, "evidence": { "release_date": [ { "value": "2026-09-10", "sources": [ "deepseek-v4-1-flash-s1", "deepseek-v4-1-flash-s2" ] } ] }, "undisclosed": [ "context_window", "parameter_count" ] }, { "id": "gpt-live-1", "model": "GPT-Live-1", "company": "OpenAI", "family": "GPT-Live", "classification": { "primary_type": "audio", "subtype": null }, "modalities": { "input": [ "audio", "text" ], "output": [ "audio", "text" ] }, "capabilities": [ "function_calling" ], "tags": [], "note": "OpenAI introduced GPT-Live-1 in the API for building natural voice experiences.", "events": [ { "type": "announcement", "date": "2026-09-10", "sources": [ "gpt-live-1-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-live-1-s1", "url": "https://openai.com/index/introducing-gpt-live-1-in-the-api/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260910170442/https://openai.com/index/introducing-gpt-live-1-in-the-api/", "retrieved": "2026-09-10" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-09-10", "sources": [ "gpt-live-1-s1" ] } ] } }, { "id": "kimi-k2-8-preview", "model": "Kimi K2.8 Preview", "company": "Moonshot AI", "family": "Kimi", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "reasoning" ], "tags": [], "note": "Rolled out in full across Kimi Code on September 11 under the unchanged model ID kimi-for-coding. Moonshot describes its performance as close to Kimi K3, with more efficient thinking than Kimi K2.7 Code, adjustable thinking effort, and a context window of up to 1M tokens on every membership tier. No weights were published.", "events": [ { "type": "announcement", "date": "2026-09-11", "sources": [ "kimi-k2-8-preview-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "kimi-k2-8-preview-s1", "url": "https://www.kimi.com/code/docs/en/kimi-code/whats-new.html", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260913163405/https://www.kimi.com/code/docs/en/kimi-code/whats-new.html", "retrieved": "2026-09-13" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-09-11", "sources": [ "kimi-k2-8-preview-s1" ] } ] } }, { "id": "gemini-3-8-live", "model": "Gemini 3.8 Live", "company": "Google DeepMind", "family": "Gemini Live", "classification": { "primary_type": "audio", "subtype": null }, "modalities": { "input": [ "text", "audio", "video" ], "output": [ "text", "audio" ] }, "capabilities": [ "reasoning", "function_calling" ], "tags": [], "note": "Google made Gemini 3.8 Live (gemini-3.8-live) generally available: an audio-to-audio model for real-time voice applications through the Live API, the default for low-latency voice agents and dialogue, with interleaved reasoning and asynchronous function calling by default.", "events": [ { "type": "announcement", "date": "2026-09-15", "sources": [ "gemini-3-8-live-s1", "gemini-3-8-live-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-3-8-live-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260916144718/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2026-09-16" }, { "id": "gemini-3-8-live-s2", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-live-gemini-3-8-live-extended-thinking/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260915174107/https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-live-gemini-3-8-live-extended-thinking/", "retrieved": "2026-09-15" }, { "id": "gemini-3-8-live-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-live-preview?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-09-15", "sources": [ "gemini-3-8-live-s1" ] } ] } }, { "id": "gemini-3-8-live-extended-thinking", "model": "Gemini 3.8 Live Extended Thinking", "company": "Google DeepMind", "family": "Gemini Live", "classification": { "primary_type": "audio", "subtype": null }, "modalities": { "input": [ "text", "audio", "video" ], "output": [ "text", "audio" ] }, "capabilities": [ "reasoning", "function_calling" ], "tags": [], "note": "Released with Gemini 3.8 Live: a high-reasoning audio-to-audio model (gemini-3.8-live-extended-thinking) that supports background reasoning during live audio interaction.", "events": [ { "type": "announcement", "date": "2026-09-15", "sources": [ "gemini-3-8-live-extended-thinking-s1", "gemini-3-8-live-extended-thinking-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-3-8-live-extended-thinking-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260916144718/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2026-09-16" }, { "id": "gemini-3-8-live-extended-thinking-s2", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-live-gemini-3-8-live-extended-thinking/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260915174107/https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-live-gemini-3-8-live-extended-thinking/", "retrieved": "2026-09-15" }, { "id": "gemini-3-8-live-extended-thinking-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.8-live-extended-thinking?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-09-15", "sources": [ "gemini-3-8-live-extended-thinking-s1" ] } ] } }, { "id": "jev", "model": "Jev", "company": "TypeSafe AI", "family": "Jev", "classification": { "primary_type": "decision", "subtype": null }, "modalities": null, "capabilities": [ "function_calling" ], "tags": [], "note": "TypeSafe AI's first model and, in its words, the first System One model. Jev takes a state and typed questions and returns every answer in parallel as calibrated probabilities with confidence scores, rather than generating text token by token. Served through TypeSafe's API as jev-1.13 and priced per input token; no weights were released. Classified `decision` under TAXONOMY §6a.", "events": [ { "type": "announcement", "date": "2026-09-15", "sources": [ "jev-s1", "jev-s2", "jev-s3" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "jev-s1", "url": "https://typesafe.ai/blog/introducing-system-one-models-and-jev", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260915192610/https://typesafe.ai/blog/introducing-system-one-models-and-jev", "retrieved": "2026-09-15" }, { "id": "jev-s2", "url": "https://docs.typesafe.ai/introduction", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260915201800/https://docs.typesafe.ai/introduction", "retrieved": "2026-09-15" }, { "id": "jev-s3", "url": "https://docs.typesafe.ai/model-jaggedness/jev-1.13", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260918003620/https://docs.typesafe.ai/model-jaggedness/jev-1.13", "retrieved": "2026-09-18" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement and official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Capabilities evidenced from the primary sources: function_calling: “Think of Jev as a frontier-intelligence function call: unstructured state in, typed probabilistic decisions out.”." }, "evidence": {} }, { "id": "grok-voice-transcribe-2-0", "model": "Grok Voice Transcribe 2.0", "company": "xAI", "family": "Grok Voice", "classification": { "primary_type": "audio", "subtype": null }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "xAI introduced Grok Voice Transcribe 2.0, available on its API as grok-voice-transcribe-2.0 alongside the earlier 1.0.", "events": [ { "type": "announcement", "date": "2026-09-18", "sources": [ "grok-voice-transcribe-2-0-s1", "grok-voice-transcribe-2-0-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "grok-voice-transcribe-2-0-s1", "url": "https://x.ai/news/grok-voice-transcribe-2", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260918175927/https://x.ai/news/grok-voice-transcribe-2", "retrieved": "2026-09-18" }, { "id": "grok-voice-transcribe-2-0-s2", "url": "https://docs.x.ai/developers/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260918184825/https://docs.x.ai/developers/release-notes", "retrieved": "2026-09-18" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`. Modalities recorded as text-to-text. The primary source states the model is text-only." }, "evidence": { "release_date": [ { "value": "2026-09-18", "sources": [ "grok-voice-transcribe-2-0-s1" ] } ] } }, { "id": "qwen-3-8-omni-flash", "model": "Qwen3.8-Omni-Flash", "company": "Alibaba Qwen", "family": "Qwen", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "agentic" ], "tags": [], "note": "Qwen launched Qwen3.8-Omni-Flash, its next-generation native omnimodal model, aimed at agentic work: planning tasks, calling tools and completing creative work across modalities. The date is the one Qwen's published research index gives (Release, 2026/09/18); the post's own page carried a September 14 date and a [draft] marker when read, and one tracker dated it September 21.", "events": [ { "type": "announcement", "date": "2026-09-18", "sources": [ "qwen-3-8-omni-flash-s1", "qwen-3-8-omni-flash-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "qwen-3-8-omni-flash-s1", "url": "https://qwen.ai/research", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260917021203/https://qwen.ai/research", "retrieved": "2026-09-17" }, { "id": "qwen-3-8-omni-flash-s2", "url": "https://qwen.ai/blog?id=qwen3.8-omni-flash", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260917230806/https://qwen.ai/blog?id=qwen3.8-omni-flash", "retrieved": "2026-09-17" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement. Not verified because the release date could not be traced: every archived snapshot of those sources predates the release — the newest is 2026-09-17, captured before the page announced it — so the archived copy does not contain the figure. A capture from on or after 2026-09-18 would settle it. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present." }, "evidence": {}, "undisclosed": [ "context_window", "parameter_count" ] }, { "id": "qwen-3-8-livetranslate", "model": "Qwen3.8-LiveTranslate", "company": "Alibaba Qwen", "family": "Qwen-LiveTranslate", "classification": { "primary_type": "audio", "subtype": null }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "multilingual" ], "tags": [], "note": "Qwen released Qwen3.8-LiveTranslate for real-time simultaneous interpretation, rebuilt on an interleave architecture; Qwen reports average lagging (LAAL) falling from 2.8 to 2.3 seconds. Dated from Qwen's published research index (Release, 2026/09/18).", "events": [ { "type": "announcement", "date": "2026-09-18", "sources": [ "qwen-3-8-livetranslate-s1", "qwen-3-8-livetranslate-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "qwen-3-8-livetranslate-s1", "url": "https://qwen.ai/research", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260917021203/https://qwen.ai/research", "retrieved": "2026-09-17" }, { "id": "qwen-3-8-livetranslate-s2", "url": "https://qwen.ai/blog?id=qwen3.8-livetranslate", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260919083220/https://qwen.ai/blog?id=qwen3.8-livetranslate", "retrieved": "2026-09-19" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. No image, audio or video capability appears anywhere in the primary sources, which labs advertise prominently when present." }, "evidence": {} }, { "id": "qwen-image-2-1", "model": "Qwen-Image-2.1", "company": "Alibaba Qwen", "family": "Qwen-Image", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": { "input": [ "text" ], "output": [ "image" ] }, "capabilities": [], "tags": [], "note": "Qwen open-sourced Qwen-Image-2.1, a unified text-to-image generation and image editing model built for efficiency, with mixed-granularity attention and prefix KV-cache reuse. Qwen states 7B parameters in its visual generation component (32 single-stream DiT layers); that is one component, not the whole model, so no total parameter count is recorded. Numbered below Qwen-Image-3.0 but released after it: this is Qwen's open-weight image line, where 3.0 is served by API.", "events": [ { "type": "announcement", "date": "2026-09-20", "sources": [ "qwen-image-2-1-s1", "qwen-image-2-1-s2" ] } ], "specifications": {}, "access": { "open_weights": true, "license": "qwen-research" }, "sources": [ { "id": "qwen-image-2-1-s1", "url": "https://qwen.ai/blog?id=qwen-image-2.1", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260920131507/https://qwen.ai/blog?id=qwen-image-2.1", "retrieved": "2026-09-20" }, { "id": "qwen-image-2-1-s2", "url": "https://huggingface.co/Qwen/Qwen-Image-2.1", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260920130007/https://huggingface.co/Qwen/Qwen-Image-2.1", "retrieved": "2026-09-20" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement and the official model card. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities from the model card's own pipeline_tag \"text-to-image\" on Qwen/Qwen-Image-2.1." }, "evidence": {} }, { "id": "grok-4-7", "model": "Grok 4.7", "company": "xAI", "family": "Grok", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "agentic", "coding", "function_calling", "structured_output" ], "tags": [], "note": "xAI made Grok 4.7 available on its API as grok-4.7, describing it as its frontier model for coding, agentic tasks and knowledge work. xAI states a 500k context window, text and image inputs with text-only output, and reasoning effort settings of low, medium, high (default) and xhigh. The date is from xAI's newsroom post of 2026-09-21; the release notes list Grok 4.7 under September without a day.", "events": [ { "type": "announcement", "date": "2026-09-21", "sources": [ "grok-4-7-s3", "grok-4-7-s1", "grok-4-7-s2" ] } ], "specifications": { "language": { "context_window": 500000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "grok-4-7-s1", "url": "https://docs.x.ai/developers/release-notes", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260918184825/https://docs.x.ai/developers/release-notes", "retrieved": "2026-09-18" }, { "id": "grok-4-7-s2", "url": "https://docs.x.ai/developers/grok-4-7", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260923093909/https://docs.x.ai/developers/grok-4-7", "retrieved": "2026-09-23" }, { "id": "grok-4-7-s3", "url": "https://x.ai/news/grok-4-7", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260921182849/https://x.ai/news/grok-4-7", "retrieved": "2026-09-21" }, { "id": "grok-4-7-xdocs", "url": "https://docs.x.ai/developers/models.md", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null }, { "id": "grok-4-7-xmodel", "url": "https://docs.x.ai/developers/models/grok-4.7", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-09-21", "sources": [ "grok-4-7-s2", "grok-4-7-s3" ] } ], "context_window": [ { "value": 500000, "sources": [ "grok-4-7-s1", "grok-4-7-s2" ] } ] }, "benchmarks": [ { "name": "HealthBench", "score": 56.7, "evaluation_type": "vendor_reported", "reported_on": "2026-09-21", "sources": [ "grok-4-7-s3" ] } ], "pricing": [ { "unit": "per_million_tokens", "rates": { "input": 2, "output": 6 }, "currency": "USD", "observed_on": "2026-09-23", "sources": [ "grok-4-7-s2" ] } ] }, { "id": "claude-opus-5-5", "model": "Claude Opus 5.5", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "reasoning", "agentic", "coding" ], "tags": [], "note": "Anthropic launched Claude Opus 5.5 (claude-opus-5-5) for long-running agentic coding and knowledge work, with a 1M token context window by default, 128k max output tokens and always-on adaptive thinking, priced at $4 / $20 per million input / output tokens against $5 / $25 for Claude Opus 5. Available on the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry.", "events": [ { "type": "announcement", "date": "2026-09-22", "sources": [ "claude-opus-5-5-s1" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-opus-5-5-s1", "url": "https://platform.claude.com/docs/en/release-notes/api", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." } }, { "id": "gpt-6-sol", "model": "GPT-6 Sol", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "reasoning" ], "tags": [], "note": "OpenAI released GPT-6 Sol (gpt-6-sol) alongside GPT-6 Luna, following GPT-6 Astra: reasoning models that accept text and image inputs and generate text through the Responses and Chat Completions APIs. OpenAI positions Sol for everyday work at lower cost than Astra. OpenAI's announcement page carries no date; the date is the one OpenAI's API changelog states.", "events": [ { "type": "announcement", "date": "2026-09-22", "sources": [ "gpt-6-sol-s1", "gpt-6-sol-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-6-sol-s1", "url": "https://developers.openai.com/api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260922191231/https://developers.openai.com/api/docs/changelog", "retrieved": "2026-09-22" }, { "id": "gpt-6-sol-s2", "url": "https://openai.com/index/introducing-gpt-6-sol-and-luna/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260922180252/https://openai.com/index/introducing-gpt-6-sol-and-luna/", "retrieved": "2026-09-22" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation and the lab's own announcement. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. The primary source states the model is text-only." }, "evidence": {}, "undisclosed": [ "parameter_count" ] }, { "id": "gpt-6-luna", "model": "GPT-6 Luna", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "structured_output", "function_calling" ], "tags": [], "note": "Released with GPT-6 Sol: a cheaper GPT-6 reasoning model optimised for fast responses and higher volume, accepting text and image inputs and generating text. OpenAI's announcement page carries no date; the date is the one OpenAI's API changelog states.", "events": [ { "type": "announcement", "date": "2026-09-22", "sources": [ "gpt-6-luna-s1", "gpt-6-luna-s2" ] } ], "specifications": { "language": { "context_window": 1050000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-6-luna-s1", "url": "https://developers.openai.com/api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260922191231/https://developers.openai.com/api/docs/changelog", "retrieved": "2026-09-22" }, { "id": "gpt-6-luna-s2", "url": "https://openai.com/index/introducing-gpt-6-sol-and-luna/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260922180252/https://openai.com/index/introducing-gpt-6-sol-and-luna/", "retrieved": "2026-09-22" }, { "id": "gpt-6-luna-odocs", "url": "https://developers.openai.com/api/docs/models/gpt-6-luna", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation and the lab's own announcement. Not verified because the release date and the context window could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": {} }, { "id": "mimo-v2-6-pro", "model": "MiMo-V2.6-Pro", "company": "Xiaomi", "family": "MiMo", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "reasoning" ], "tags": [], "note": "Xiaomi's most capable model at release: an omni-modal, trillion-parameter reasoning model for complex projects, long-horizon tasks, cybersecurity and research. Weights are published as MiMo-V2.6-Pro-RL, with a 1,048,576-token context in the config. Xiaomi also serves it as mimo-v2.6-pro-ultraspeed, the same model up to 20x faster — a serving variant, recorded here rather than as a separate model.", "events": [ { "type": "announcement", "date": "2026-09-22", "sources": [ "mimo-v2-6-pro-s1", "mimo-v2-6-pro-s2" ] } ], "specifications": { "language": { "context_window": 1048576, "parameter_count": null } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "mimo-v2-6-pro-s1", "url": "https://mimo.mi.com/docs/en-US/updates/model", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260922031526/https://mimo.mi.com/docs/en-US/updates/model", "retrieved": "2026-09-22" }, { "id": "mimo-v2-6-pro-s2", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260921231812/https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL", "retrieved": "2026-09-21" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-09-22", "sources": [ "mimo-v2-6-pro-s1" ] } ], "context_window": [ { "value": 1048576, "sources": [ "mimo-v2-6-pro-s1", "mimo-v2-6-pro-s2" ] } ] } }, { "id": "mimo-v2-6-flash", "model": "MiMo-V2.6-Flash", "company": "Xiaomi", "family": "MiMo", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "reasoning" ], "tags": [], "note": "Released with MiMo-V2.6-Pro: a full-modality, low-cost reasoning model for high-frequency calls and large-scale professional workloads. Weights are published as MiMo-V2.6-Flash-RL, with a 1,048,576-token context in the config.", "events": [ { "type": "announcement", "date": "2026-09-22", "sources": [ "mimo-v2-6-flash-s1", "mimo-v2-6-flash-s2" ] } ], "specifications": { "language": { "context_window": 1048576, "parameter_count": null } }, "access": { "open_weights": true, "license": "MIT" }, "sources": [ { "id": "mimo-v2-6-flash-s1", "url": "https://mimo.mi.com/docs/en-US/updates/model", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260922031526/https://mimo.mi.com/docs/en-US/updates/model", "retrieved": "2026-09-22" }, { "id": "mimo-v2-6-flash-s2", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260921234213/https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL", "retrieved": "2026-09-21" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-09-22", "sources": [ "mimo-v2-6-flash-s1" ] } ], "context_window": [ { "value": 1048576, "sources": [ "mimo-v2-6-flash-s1", "mimo-v2-6-flash-s2" ] } ] } }, { "id": "gemini-3-8-flash-tts", "model": "Gemini 3.8 Flash TTS", "company": "Google DeepMind", "family": "Gemini TTS", "classification": { "primary_type": "audio", "subtype": null }, "modalities": { "input": [ "text" ], "output": [ "audio" ] }, "capabilities": [], "tags": [], "note": "Google made Gemini 3.8 Flash TTS (gemini-3.8-flash-tts) generally available as its flagship next-generation text-to-speech model, alongside a Voices endpoint in the Gemini API.", "events": [ { "type": "announcement", "date": "2026-09-22", "sources": [ "gemini-3-8-flash-tts-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-3-8-flash-tts-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260921201341/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2026-09-21" }, { "id": "gemini-3-8-flash-tts-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.8-flash-tts?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be traced: every archived snapshot of those sources predates the release — the newest is 2026-09-21, captured before the page announced it — so the archived copy does not contain the figure. A capture from on or after 2026-09-22 would settle it." }, "evidence": {} }, { "id": "gemini-3-8-flash-lite-tts", "model": "Gemini 3.8 Flash-Lite TTS", "company": "Google DeepMind", "family": "Gemini TTS", "classification": { "primary_type": "audio", "subtype": null }, "modalities": { "input": [ "text" ], "output": [ "audio" ] }, "capabilities": [], "tags": [], "note": "Released with Gemini 3.8 Flash TTS: the lighter of Google's two next-generation text-to-speech models.", "events": [ { "type": "announcement", "date": "2026-09-22", "sources": [ "gemini-3-8-flash-lite-tts-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-3-8-flash-lite-tts-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260921201341/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2026-09-21" }, { "id": "gemini-3-8-flash-lite-tts-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-tts-preview?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date could not be traced: every archived snapshot of those sources predates the release — the newest is 2026-09-21, captured before the page announced it — so the archived copy does not contain the figure. A capture from on or after 2026-09-22 would settle it." }, "evidence": {} }, { "id": "nemotron-3-diarization", "model": "Nemotron 3 Diarization", "company": "NVIDIA", "family": "Nemotron Diarization", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "An open-weights speaker diarization model of about 100M parameters: from 16 kHz audio it reports who spoke when, for up to eight speakers including overlapping speech, in streaming or offline mode. Its Hugging Face repository was created on September 1; NVIDIA announced it on September 23.", "events": [ { "type": "announcement", "date": "2026-09-23", "sources": [ "nemotron-3-diarization-s1", "nemotron-3-diarization-s2" ] } ], "specifications": {}, "access": { "open_weights": true, "license": "OpenMDW License 1.1" }, "sources": [ { "id": "nemotron-3-diarization-s1", "url": "https://huggingface.co/blog/nvidia/nemotron-diarization", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260923141752/https://huggingface.co/blog/nvidia/nemotron-diarization", "retrieved": "2026-09-23" }, { "id": "nemotron-3-diarization-s2", "url": "https://huggingface.co/nvidia/Nemotron-3-Diarization", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20260923185654/https://huggingface.co/nvidia/Nemotron-3-Diarization", "retrieved": "2026-09-23" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-09-23", "sources": [ "nemotron-3-diarization-s1", "nemotron-3-diarization-s2" ] } ] } }, { "id": "minimax-m3-1-flash-preview", "model": "MiniMax M3.1-Flash-Preview", "company": "MiniMax", "family": "MiniMax", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "reasoning" ], "tags": [], "note": "The first M3.1-branded model anyone could use, launched on September 27 inside MiniMax Code and the Token Plan subscription rather than pay-as-you-go. MiniMax's documentation calls it a multimodal coding model with a 1M-token context window and tunable thinking depth. The date is that of MiniMax's own launch post; its newsroom carried no announcement.", "events": [ { "type": "announcement", "date": "2026-09-27", "sources": [ "minimax-m3-1-flash-preview-s1", "minimax-m3-1-flash-preview-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "minimax-m3-1-flash-preview-s1", "url": "https://x.com/MiniMax_AI/status/2104256406786547800", "type": "official_announcement", "authority": "primary", "archived_url": null, "retrieved": null }, { "id": "minimax-m3-1-flash-preview-s2", "url": "https://platform.minimax.io/docs/guides/models-intro", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260930092243/https://platform.minimax.io/docs/guides/models-intro", "retrieved": "2026-09-30" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement and official documentation. Not verified because the release date could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from." }, "evidence": {} }, { "id": "claude-sonnet-5-5", "model": "Claude Sonnet 5.5", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "reasoning" ], "tags": [], "note": "Anthropic launched Claude Sonnet 5.5 (claude-sonnet-5-5) on the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry. Anthropic's models table gives it a 1M-token context window, 128K maximum output, adaptive thinking, and pricing of $2 / $10 per million input / output tokens.", "events": [ { "type": "announcement", "date": "2026-09-28", "sources": [ "claude-sonnet-5-5-s1", "claude-sonnet-5-5-s2" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-sonnet-5-5-s1", "url": "https://platform.claude.com/docs/en/release-notes/api", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null }, { "id": "claude-sonnet-5-5-s2", "url": "https://platform.claude.com/docs/en/about-claude/models", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." } }, { "id": "gpt-6-1-sol", "model": "GPT-6.1 Sol", "company": "OpenAI", "family": "GPT", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [ "coding", "structured_output", "function_calling", "reasoning" ], "tags": [], "note": "OpenAI released GPT-6.1 Sol (gpt-6.1-sol) in the Responses and Chat Completions APIs, for complex coding and professional work at a lower cost than GPT-6 Astra.", "events": [ { "type": "announcement", "date": "2026-09-29", "sources": [ "gpt-6-1-sol-s1" ] } ], "specifications": { "language": { "context_window": 1050000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gpt-6-1-sol-s1", "url": "https://developers.openai.com/api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260928233209/https://developers.openai.com/api/docs/changelog", "retrieved": "2026-09-28" }, { "id": "gpt-6-1-sol-odocs", "url": "https://developers.openai.com/api/docs/models/gpt-6.1-sol", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to official documentation. Not verified because the release date and the context window could not be traced: every archived snapshot of those sources predates the release — the newest is 2026-09-28, captured before the page announced it — so the archived copy does not contain the figure. A capture from on or after 2026-09-29 would settle it." }, "evidence": {} }, { "id": "gemini-4-argon", "model": "Gemini 4 Argon", "company": "Google DeepMind", "family": "Gemini", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "capabilities": [ "reasoning", "function_calling", "structured_output", "agentic" ], "tags": [], "note": "Google's first Gemini 4 model, announced as its most capable yet. At announcement it was rolling out only to trusted cyber defenders through Google DeepMind's Fairwind Program, with paid API customers and Google AI Ultra subscribers to follow. Google states a 1M-token output limit; it does not state the input context window, and no API model page existed at announcement.", "events": [ { "type": "announcement", "date": "2026-09-30", "sources": [ "gemini-4-argon-s1", "gemini-4-argon-s2" ] }, { "type": "api_availability", "date": "2026-10-13", "sources": [ "gemini-4-argon-s3" ] }, { "type": "public_availability", "date": "2026-10-13", "sources": [ "gemini-4-argon-s4" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "gemini-4-argon-s1", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20260930200646/https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/", "retrieved": "2026-09-30" }, { "id": "gemini-4-argon-s2", "url": "https://deepmind.google/fairwind-program/", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20260929123717/https://deepmind.google/fairwind-program/", "retrieved": "2026-09-29" }, { "id": "gemini-4-argon-s3", "url": "https://blog.google/innovation-and-ai/technology/developers-tools/build-with-gemini-4-argon/", "type": "official_announcement", "authority": "primary", "archived_url": null, "retrieved": null }, { "id": "gemini-4-argon-s4", "url": "https://blog.google/innovation-and-ai/products/gemini-app/gemini-4-argon-availability/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261013160816/https://blog.google/innovation-and-ai/products/gemini-app/gemini-4-argon-availability/", "retrieved": "2026-10-13" }, { "id": "gemini-4-argon-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-4-argon?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-09-30", "sources": [ "gemini-4-argon-s1" ] } ] } }, { "id": "mai-transcribe-2-streaming", "model": "MAI-Transcribe-2-Streaming", "company": "Microsoft", "family": "MAI-Transcribe", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "Microsoft AI's first streaming transcription model: low-latency, real-time transcripts in 60 languages with continuous automatic language detection. Announced alongside MAI-Voice-2.1 and MAI-Voice-2.1-Flash.", "events": [ { "type": "announcement", "date": "2026-10-01", "sources": [ "mai-transcribe-2-streaming-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mai-transcribe-2-streaming-s1", "url": "https://microsoft.ai/news/our-first-streaming-transcription-model/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261001180811/https://microsoft.ai/news/our-first-streaming-transcription-model/", "retrieved": "2026-10-01" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-10-01", "sources": [ "mai-transcribe-2-streaming-s1" ] } ] } }, { "id": "mai-voice-2-1", "model": "MAI-Voice-2.1", "company": "Microsoft", "family": "MAI-Voice", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "Microsoft AI's multilingual text-to-speech model, covering 23 languages and 26 locales. Announced alongside MAI-Transcribe-2-Streaming.", "events": [ { "type": "announcement", "date": "2026-10-01", "sources": [ "mai-voice-2-1-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mai-voice-2-1-s1", "url": "https://microsoft.ai/news/our-first-streaming-transcription-model/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261001180811/https://microsoft.ai/news/our-first-streaming-transcription-model/", "retrieved": "2026-10-01" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-10-01", "sources": [ "mai-voice-2-1-s1" ] } ] } }, { "id": "mai-voice-2-1-flash", "model": "MAI-Voice-2.1-Flash", "company": "Microsoft", "family": "MAI-Voice", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "The lower-cost tier of MAI-Voice-2.1 for high-volume, latency-sensitive speech, with the same language coverage. Announced alongside MAI-Transcribe-2-Streaming.", "events": [ { "type": "announcement", "date": "2026-10-01", "sources": [ "mai-voice-2-1-flash-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mai-voice-2-1-flash-s1", "url": "https://microsoft.ai/news/our-first-streaming-transcription-model/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261001180811/https://microsoft.ai/news/our-first-streaming-transcription-model/", "retrieved": "2026-10-01" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-10-01", "sources": [ "mai-voice-2-1-flash-s1" ] } ] } }, { "id": "strands-decider-2b", "model": "Strands Decider 2B", "company": "Amazon", "family": "Strands Decider", "classification": { "primary_type": "decision", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "An open decision model from AWS's Strands Labs, citing TypeSafe AI's Jev as the model that drew attention to the class. Given a state and typed questions supplied with the request (a choice, a yes/no, a score), it returns each answer with a calibrated confidence from a small readout head, not by generating text. 1.9B parameters: a LoRA adapter on Qwen3.5-2B-Base plus that head, released with its training data and scripts.", "events": [ { "type": "announcement", "date": "2026-10-01", "sources": [ "strands-decider-2b-s1", "strands-decider-2b-s2", "strands-decider-2b-s3" ] } ], "specifications": {}, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "strands-decider-2b-s1", "url": "https://strandsagents.com/blog/introducing-strands-decider/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261001185944/https://strandsagents.com/blog/introducing-strands-decider/", "retrieved": "2026-10-01" }, { "id": "strands-decider-2b-s2", "url": "https://github.com/strands-labs/strands-decider", "type": "official_repository", "authority": "primary", "archived_url": "https://web.archive.org/web/20261001231154/https://github.com/strands-labs/strands-decider", "retrieved": "2026-10-01" }, { "id": "strands-decider-2b-s3", "url": "https://huggingface.co/StrandsAgents/strands-decider-2B-hobson-v19", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20261002064735/https://huggingface.co/StrandsAgents/strands-decider-2B-hobson-v19", "retrieved": "2026-10-02" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-10-01", "sources": [ "strands-decider-2b-s3" ] } ] } }, { "id": "clef", "model": "Clef", "company": "Cloudflare", "family": "Clef", "classification": { "primary_type": "decision", "subtype": null }, "modalities": null, "capabilities": [ "agentic", "video" ], "tags": [], "note": "Cloudflare's open decision model, served on Workers AI. A frozen Qwen3.8-27B with low-rank adapters and a routing head: questions and answer types arrive with each request, and it scores the valid answers in one prefill pass, returning typed answers with probabilities rather than generating text.", "events": [ { "type": "announcement", "date": "2026-10-01", "sources": [ "clef-s1", "clef-s2" ] } ], "specifications": {}, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "clef-s1", "url": "https://blog.cloudflare.com/clef-decision-models/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261001153518/https://blog.cloudflare.com/clef-decision-models/", "retrieved": "2026-10-01" }, { "id": "clef-s2", "url": "https://huggingface.co/Cloudflare/clef", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20261001183535/https://huggingface.co/Cloudflare/clef", "retrieved": "2026-10-01" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`. Capabilities evidenced from the primary sources: agentic: “How fine-tuning can extend the capabilities of Clef We heard a lot of internal use cases that required fine-tuning our Clef model to be built into our agentic workflows at Cloudfla” | video: “co/Cloudflare/clef Clef Model Files Usage Jev / SystemOne API Images and video Input format Results Decision Index Workflow evals License Clef Announcement: Clef decision models on”." }, "evidence": { "release_date": [ { "value": "2026-10-01", "sources": [ "clef-s1", "clef-s2" ] } ] } }, { "id": "clef-flash", "model": "Clef-flash", "company": "Cloudflare", "family": "Clef", "classification": { "primary_type": "decision", "subtype": null }, "modalities": null, "capabilities": [ "video" ], "tags": [], "note": "The faster sibling of Clef for latency-critical decisions, built the same way on Qwen3.5-9B.", "events": [ { "type": "announcement", "date": "2026-10-01", "sources": [ "clef-flash-s1", "clef-flash-s2" ] } ], "specifications": {}, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "clef-flash-s1", "url": "https://blog.cloudflare.com/clef-decision-models/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261001153518/https://blog.cloudflare.com/clef-decision-models/", "retrieved": "2026-10-01" }, { "id": "clef-flash-s2", "url": "https://huggingface.co/Cloudflare/clef-flash", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20261001192910/https://huggingface.co/Cloudflare/clef-flash", "retrieved": "2026-10-01" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`. Capabilities evidenced from the primary sources: video: “Clef-Flash Model Files Usage Jev / SystemOne API Images and video Input format Results Decision Index Workflow evals License Clef-Flash Announcement: Clef decision models on the Cl”." }, "evidence": { "release_date": [ { "value": "2026-10-01", "sources": [ "clef-flash-s1" ] } ] } }, { "id": "pplx-decider-v1-27b", "model": "pplx-decider-v1-27b", "company": "Perplexity", "family": "pplx-decider", "classification": { "primary_type": "decision", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "The open-weights model behind Perplexity's Decisions API. Given a state of text, JSON or images and 1 to 128 typed questions per request (yes/no, choice, score), it returns a probability for each answer and does not write text. Perplexity's changelog later names pplx-decider-v1.1-27b as its replacement.", "events": [ { "type": "announcement", "date": "2026-10-01", "sources": [ "pplx-decider-v1-27b-s1", "pplx-decider-v1-27b-s2", "pplx-decider-v1-27b-s3" ] } ], "specifications": {}, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "pplx-decider-v1-27b-s1", "url": "https://x.com/perplexitydevs/status/2105725598882832414", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261002012621/https://x.com/perplexitydevs/status/2105725598882832414", "retrieved": "2026-10-02" }, { "id": "pplx-decider-v1-27b-s2", "url": "https://docs.perplexity.ai/docs/decisions/quickstart", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20261001185909/https://docs.perplexity.ai/docs/decisions/quickstart", "retrieved": "2026-10-01" }, { "id": "pplx-decider-v1-27b-s3", "url": "https://huggingface.co/perplexity-ai/pplx-decider-v1-27b", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20261001210107/https://huggingface.co/perplexity-ai/pplx-decider-v1-27b", "retrieved": "2026-10-01" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-10-01", "sources": [ "pplx-decider-v1-27b-s1", "pplx-decider-v1-27b-s3" ] } ] } }, { "id": "nova-2-5-sonic", "model": "Nova 2.5 Sonic", "company": "Amazon", "family": "Nova Sonic", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "Amazon's speech-to-speech model for real-time voice agents, generally available in Amazon Bedrock in four regions. Amazon describes improved reasoning, instruction following, tool calling and latency over Nova 2 Sonic, which this dataset does not yet record.", "events": [ { "type": "announcement", "date": "2026-10-05", "sources": [ "nova-2-5-sonic-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "nova-2-5-sonic-s1", "url": "https://aws.amazon.com/about-aws/whats-new/2026/10/amazon-nova-2.5-sonic/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261006012046/https://aws.amazon.com/about-aws/whats-new/2026/10/amazon-nova-2.5-Sonic/", "retrieved": "2026-10-06" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-10-05", "sources": [ "nova-2-5-sonic-s1" ] } ] } }, { "id": "nano-banana-2-1", "model": "Nano Banana 2.1", "company": "Google DeepMind", "family": "Nano Banana", "classification": { "primary_type": "image_generation", "subtype": null }, "modalities": { "input": [ "text", "image", "video" ], "output": [ "image", "text" ] }, "capabilities": [ "reasoning", "image_generation" ], "tags": [], "note": "Google's high-efficiency image generation and conversational editing model, released generally available as gemini-nano-banana-2.1. Google describes it as an update to Nano Banana 2 (gemini-3.1-flash-image).", "events": [ { "type": "announcement", "date": "2026-10-06", "sources": [ "nano-banana-2-1-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "nano-banana-2-1-s1", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "official_documentation", "authority": "primary", "archived_url": "https://web.archive.org/web/20261006162120/https://ai.google.dev/gemini-api/docs/changelog", "retrieved": "2026-10-06" }, { "id": "nano-banana-2-1-gdocs", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-nano-banana-2.1?hl=en", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-10-06", "sources": [ "nano-banana-2-1-s1" ] } ] } }, { "id": "embeddinggemma-2", "model": "EmbeddingGemma 2", "company": "Google DeepMind", "family": "Gemma", "classification": { "primary_type": "language", "subtype": "embedding" }, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": [ "video" ], "tags": [], "note": "Google's open-weights multimodal embedding model: 740M parameters mapping text, images, video frames and audio into one vector space, built to run on device. A text-only subset loads alone.", "events": [ { "type": "announcement", "date": "2026-10-06", "sources": [ "embeddinggemma-2-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 740000000 } }, "access": { "open_weights": true, "license": null }, "sources": [ { "id": "embeddinggemma-2-s1", "url": "https://developers.googleblog.com/google-ai-edge-with-embeddinggemma-2/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261006172426/https://developers.googleblog.com/google-ai-edge-with-embeddinggemma-2/", "retrieved": "2026-10-06" } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Cited to the lab's own announcement. Not verified because the release date and the parameter count could not be found in any cited primary source — the value may be correct, but this dataset cannot yet show where it came from. Modalities recorded as text-to-text. The primary source states the model is text-only. Capabilities evidenced from the primary sources: video: “6, 2026 Google AI Edge Team Android ML Team Share Facebook Twitter LinkedIn Mail Today, Google DeepMind launched EmbeddingGemma 2 , a best-in-class for its size, open-weight, multi”." }, "evidence": {}, "undisclosed": [ "context_window", "license" ] }, { "id": "mistral-large-4", "model": "Mistral Large 4", "company": "Mistral AI", "family": "Mistral", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "coding", "agentic" ], "tags": [], "note": "A natively multimodal mixture-of-experts model with 1 trillion total and 52 billion active parameters, released as a public preview API on Mistral Studio. Mistral said it would release the weights by the end of the month; none were public at announcement.", "events": [ { "type": "announcement", "date": "2026-10-06", "sources": [ "mistral-large-4-s1" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 1000000000000 } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "mistral-large-4-s1", "url": "https://mistral.ai/news/mistral-large-4/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261006132643/https://mistral.ai/news/mistral-large-4/", "retrieved": "2026-10-06" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`. Capabilities evidenced from the primary sources: coding: “Human Evaluation We ran an internal evaluation in which expert annotators across coding, computer-aided design (CAD), finance, mathematics and physics compared Mistral Large 4 with” | agentic: “Mistral Large 4 New Open-weight hybrid instruct-and-reasoning MoE with multimodal input; unifies instruction, reasoning, and agentic capabilities in a single model, state-of-the-ar”." }, "evidence": { "release_date": [ { "value": "2026-10-06", "sources": [ "mistral-large-4-s1" ] } ], "parameter_count": [ { "value": 1000000000000, "sources": [ "mistral-large-4-s1" ] } ] }, "benchmarks": [ { "name": "Terminal-Bench", "score": 28.3, "evaluation_type": "vendor_reported", "reported_on": "2026-10-06", "sources": [ "mistral-large-4-s1" ] } ], "undisclosed": [ "context_window" ] }, { "id": "claude-haiku-5-5", "model": "Claude Haiku 5.5", "company": "Anthropic", "family": "Claude", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "reasoning" ], "tags": [], "note": "Anthropic's model tuned for high-volume, latency-sensitive work, released on the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry with a 1M-token context window, 128k max output tokens and adaptive thinking. Its API id is claude-haiku-5-5.", "events": [ { "type": "announcement", "date": "2026-10-07", "sources": [ "claude-haiku-5-5-s1" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "claude-haiku-5-5-s1", "url": "https://platform.claude.com/docs/en/release-notes/api", "type": "official_documentation", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." } }, { "id": "d1-3b", "model": "d1-3B", "company": "Liquid AI", "family": "d1", "classification": { "primary_type": "decision", "subtype": null }, "modalities": null, "capabilities": [ "vision" ], "tags": [], "note": "An open-weights decision model built from LFM2.5-VL-3B. It takes text and images with typed questions supplied at request time (yes/no, choice, score) and answers in a single forward pass with probabilities and a confidence, generating no tokens.", "events": [ { "type": "announcement", "date": "2026-10-07", "sources": [ "d1-3b-s1", "d1-3b-s2" ] } ], "specifications": {}, "access": { "open_weights": true, "license": "LFM-1.0" }, "sources": [ { "id": "d1-3b-s1", "url": "https://huggingface.co/blog/LiquidAI/open-d1", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261007183512/https://huggingface.co/blog/LiquidAI/open-d1", "retrieved": "2026-10-07" }, { "id": "d1-3b-s2", "url": "https://huggingface.co/LiquidAI/d1-3B", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20261007171228/https://huggingface.co/LiquidAI/d1-3B", "retrieved": "2026-10-07" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`. Capabilities evidenced from the primary sources: vision: “6B • Updated about 2 hours ago • 28 • 21 Spaces mentioned in this article 1 Running on L4 3 System One Arcade 🟣 3 Ten camera, drawing and text demos of d1-3B More from this author”." }, "evidence": { "release_date": [ { "value": "2026-10-07", "sources": [ "d1-3b-s2" ] } ] } }, { "id": "d1-omni-600m", "model": "d1-omni-600M", "company": "Liquid AI", "family": "d1", "classification": { "primary_type": "decision", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "The experimental small sibling of d1-3B, about 600M parameters, built from LFM2.5-Encoder-350M. It answers typed questions about text with images or audio in a single forward pass.", "events": [ { "type": "announcement", "date": "2026-10-07", "sources": [ "d1-omni-600m-s1", "d1-omni-600m-s2" ] } ], "specifications": {}, "access": { "open_weights": true, "license": "LFM-1.0" }, "sources": [ { "id": "d1-omni-600m-s1", "url": "https://huggingface.co/blog/LiquidAI/open-d1", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261007183512/https://huggingface.co/blog/LiquidAI/open-d1", "retrieved": "2026-10-07" }, { "id": "d1-omni-600m-s2", "url": "https://huggingface.co/LiquidAI/d1-omni-600M", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20261007171456/https://huggingface.co/LiquidAI/d1-omni-600M", "retrieved": "2026-10-07" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." }, "evidence": { "release_date": [ { "value": "2026-10-07", "sources": [ "d1-omni-600m-s2" ] } ] } }, { "id": "microsoft-decision-1", "model": "Microsoft-Decision-1", "company": "Microsoft", "family": "Microsoft-Decision", "classification": { "primary_type": "decision", "subtype": null }, "modalities": null, "capabilities": [ "multilingual" ], "tags": [], "note": "A decision model from Microsoft's Office of the CTO, in public preview in Microsoft Foundry. Post-trained from Qwen3.5-9B for single-pass decision scoring: given yes/no, multiple-choice, rating or rubric options with a request, it returns a calibrated probability for each option rather than generated text. Microsoft says later versions will be rebased on OpenAI and MAI models. No weights were published.", "events": [ { "type": "announcement", "date": "2026-10-09", "sources": [ "microsoft-decision-1-s1", "microsoft-decision-1-s2" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "microsoft-decision-1-s1", "url": "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/introducing-microsoft-decision-1-in-microsoft-foundry-for-decision-and-classific/4562742", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261009193331/https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/introducing-microsoft-decision-1-in-microsoft-foundry-for-decision-and-classific/4562742", "retrieved": "2026-10-09" }, { "id": "microsoft-decision-1-s2", "url": "https://commandline.microsoft.com/microsoft-decision-1-model-foundry/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261009183947/https://commandline.microsoft.com/microsoft-decision-1-model-foundry/", "retrieved": "2026-10-09" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`. Capabilities evidenced from the primary sources: multilingual: “That’s why we evaluated Microsoft-Decision-1 across dozens of benchmarks kept blinded from training, spanning routing, ranking, long context, multilingual and out-of-distribution t”." }, "evidence": { "release_date": [ { "value": "2026-10-09", "sources": [ "microsoft-decision-1-s2" ] } ] } }, { "id": "clef-omni", "model": "Clef-omni", "company": "Cloudflare", "family": "Clef", "classification": { "primary_type": "decision", "subtype": null }, "modalities": null, "capabilities": [ "video", "vision" ], "tags": [], "note": "A multimodal Clef that answers typed questions about audio, video, images and text together. Announced with a faster Clef and a cheaper Clef-flash.", "events": [ { "type": "announcement", "date": "2026-10-09", "sources": [ "clef-omni-s1", "clef-omni-s2" ] } ], "specifications": {}, "access": { "open_weights": true, "license": "Apache-2.0" }, "sources": [ { "id": "clef-omni-s1", "url": "https://blog.cloudflare.com/clef-faster-cheaper-multimodal/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261009182831/https://blog.cloudflare.com/clef-faster-cheaper-multimodal/", "retrieved": "2026-10-09" }, { "id": "clef-omni-s2", "url": "https://huggingface.co/Cloudflare/clef-omni", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20261009185203/https://huggingface.co/Cloudflare/clef-omni", "retrieved": "2026-10-09" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`. Capabilities evidenced from the primary sources: video: “Today, we’re releasing Clef-omni, which takes in audio and video input alongside text and image.” | vision: “Clef-omni takes audio, video, image, and text input Our new Clef-omni model is able to take audio, video, image, and text input.”." }, "evidence": { "release_date": [ { "value": "2026-10-09", "sources": [ "clef-omni-s1" ] } ] } }, { "id": "muse-wonder", "model": "Muse Wonder", "company": "Meta AI", "family": "Muse", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "agentic", "coding" ], "tags": [], "note": "Meta Superintelligence Labs' first frontier model, announced with Muse Wonder Full as its most capable models, built for personal and professional agentic tasks. Available at announcement through the Meta Model API and Muse Code. No weights were published.", "events": [ { "type": "announcement", "date": "2026-10-12", "sources": [ "muse-wonder-s1", "muse-wonder-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "muse-wonder-s1", "url": "https://about.fb.com/news/2026/10/introducing-muse-wonder-models/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261012223110/https://about.fb.com/news/2026/10/introducing-muse-wonder-models/", "retrieved": "2026-10-12" }, { "id": "muse-wonder-s2", "url": "https://research.meta.ai/blog/muse-wonder", "type": "official_announcement", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there. Capabilities evidenced from the primary sources: agentic: “mp4 Takeaways Meta is introducing Muse Wonder and Muse Wonder Full, our most capable models to date, built to deliver state-of-the-art performance on personal and professional agen” | coding: “Muse Wonder and Muse Wonder Full bring that capability to expert domains including law, finance, and health, along with frontier coding performance for people building everything f”." }, "evidence": { "release_date": [ { "value": "2026-10-12", "sources": [ "muse-wonder-s1" ] } ] } }, { "id": "muse-wonder-full", "model": "Muse Wonder Full", "company": "Meta AI", "family": "Muse", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [ "agentic", "coding" ], "tags": [], "note": "Announced alongside Muse Wonder as a separately named model. At announcement it was available only to Muse Code subscribers, not through the Meta Model API. Meta's posts do not say how it differs from Muse Wonder.", "events": [ { "type": "announcement", "date": "2026-10-12", "sources": [ "muse-wonder-full-s1", "muse-wonder-full-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": null } }, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "muse-wonder-full-s1", "url": "https://about.fb.com/news/2026/10/introducing-muse-wonder-models/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261012223110/https://about.fb.com/news/2026/10/introducing-muse-wonder-models/", "retrieved": "2026-10-12" }, { "id": "muse-wonder-full-s2", "url": "https://research.meta.ai/blog/muse-wonder", "type": "official_announcement", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there. Capabilities evidenced from the primary sources: agentic: “mp4 Takeaways Meta is introducing Muse Wonder and Muse Wonder Full, our most capable models to date, built to deliver state-of-the-art performance on personal and professional agen” | coding: “Muse Wonder and Muse Wonder Full bring that capability to expert domains including law, finance, and health, along with frontier coding performance for people building everything f”." }, "evidence": { "release_date": [ { "value": "2026-10-12", "sources": [ "muse-wonder-full-s1" ] } ] } }, { "id": "grok-voice-3-0", "model": "Grok Voice 3.0", "company": "xAI", "family": "Grok Voice", "classification": { "primary_type": "audio", "subtype": null }, "modalities": null, "capabilities": [], "tags": [], "note": "xAI's speech-to-speech model, available in the API at launch at the same price as its predecessor. xAI says it already powers conversations across Tesla and SpaceX.", "events": [ { "type": "announcement", "date": "2026-10-14", "sources": [ "grok-voice-3-0-s1" ] } ], "specifications": {}, "access": { "open_weights": false, "license": null }, "sources": [ { "id": "grok-voice-3-0-s1", "url": "https://x.ai/news/grok-voice-3", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261014164929/https://x.ai/news/grok-voice-3", "retrieved": "2026-10-14" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement, and every value it asserts was found there." }, "evidence": { "release_date": [ { "value": "2026-10-14", "sources": [ "grok-voice-3-0-s1" ] } ] } }, { "id": "nemotron-3-5-super-vl", "model": "Nemotron 3.5 Super VL", "company": "NVIDIA", "family": "Nemotron", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "An open-weights multimodal mixture-of-experts model, 120B total and 12B active parameters, for text, image, document and video understanding with reasoning and tool use. NVIDIA says it was distilled from Nemotron 3 Ultra; base, post-trained and NVFP4 checkpoints were released.", "events": [ { "type": "announcement", "date": "2026-10-15", "sources": [ "nemotron-3-5-super-vl-s1", "nemotron-3-5-super-vl-s2", "nemotron-3-5-super-vl-s3" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 120000000000 } }, "access": { "open_weights": true, "license": "OpenMDW License 1.1" }, "sources": [ { "id": "nemotron-3-5-super-vl-s1", "url": "https://developer.nvidia.com/blog/nvidia-nemotron-3-5-super-vl-powers-specialized-agents-with-multimodal-reasoning/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261015155204/https://developer.nvidia.com/blog/nvidia-nemotron-3-5-super-vl-powers-specialized-agents-with-multimodal-reasoning/", "retrieved": "2026-10-15" }, { "id": "nemotron-3-5-super-vl-s2", "url": "https://blogs.nvidia.com/blog/nemotron-3-5-super-vl/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261015191910/https://blogs.nvidia.com/blog/nemotron-3-5-super-vl/", "retrieved": "2026-10-15" }, { "id": "nemotron-3-5-super-vl-s3", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20261015145418/https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16", "retrieved": "2026-10-15" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement and the official model card, and every value it asserts was found there. Modalities from the model card's own pipeline_tag \"image-text-to-text\" on nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16." }, "evidence": { "release_date": [ { "value": "2026-10-15", "sources": [ "nemotron-3-5-super-vl-s1", "nemotron-3-5-super-vl-s2", "nemotron-3-5-super-vl-s3" ] } ], "parameter_count": [ { "value": 120000000000, "sources": [ "nemotron-3-5-super-vl-s1", "nemotron-3-5-super-vl-s2", "nemotron-3-5-super-vl-s3" ] } ] } }, { "id": "nemotron-3-5-bolt-74b-a9b", "model": "Nemotron 3.5 Bolt 74B-A9B", "company": "NVIDIA", "family": "Nemotron", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": [], "tags": [], "note": "An open vision-language mixture-of-experts model, 74B total and 9B active parameters, built for local agentic work and packed into 24GB with 2-bit quantization-aware training. NVIDIA announced a smaller Bolt E4B alongside it for October 23.", "events": [ { "type": "announcement", "date": "2026-10-15", "sources": [ "nemotron-3-5-bolt-74b-a9b-s1", "nemotron-3-5-bolt-74b-a9b-s2" ] } ], "specifications": { "language": { "context_window": null, "parameter_count": 74000000000 } }, "access": { "open_weights": true, "license": "OpenMDW License 1.1" }, "sources": [ { "id": "nemotron-3-5-bolt-74b-a9b-s1", "url": "https://blogs.nvidia.com/blog/nemotron-3-5-super-vl/", "type": "official_announcement", "authority": "primary", "archived_url": "https://web.archive.org/web/20261015191910/https://blogs.nvidia.com/blog/nemotron-3-5-super-vl/", "retrieved": "2026-10-15" }, { "id": "nemotron-3-5-bolt-74b-a9b-s2", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16", "type": "official_model_card", "authority": "primary", "archived_url": "https://web.archive.org/web/20261015150640/https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16", "retrieved": "2026-10-15" } ], "provenance": { "status": "verified", "confidence": 90, "reason": "Cited to the lab's own announcement and the official model card, and every value it asserts was found there. Modalities from the model card's own pipeline_tag \"image-text-to-text\" on nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16." }, "evidence": { "release_date": [ { "value": "2026-10-15", "sources": [ "nemotron-3-5-bolt-74b-a9b-s1", "nemotron-3-5-bolt-74b-a9b-s2" ] } ], "parameter_count": [ { "value": 74000000000, "sources": [ "nemotron-3-5-bolt-74b-a9b-s1", "nemotron-3-5-bolt-74b-a9b-s2" ] } ] }, "undisclosed": [ "context_window" ] }, { "id": "muse-spark-open", "model": "Muse Spark Open", "company": "Meta AI", "family": "Muse", "classification": { "primary_type": "language", "subtype": "llm" }, "modalities": null, "capabilities": [], "tags": [], "note": "The open-weights version of Muse Spark, with additional safety post-training for open release: 816B total and 42B active parameters, natively multimodal, with a 1M-token context window. Weights are published by meta-models on Hugging Face under the Muse Model License, which limits commercial use.", "events": [ { "type": "announcement", "date": "2026-10-15", "sources": [ "muse-spark-open-s1", "muse-spark-open-s2" ] } ], "specifications": { "language": { "context_window": 1000000, "parameter_count": 816000000000 } }, "access": { "open_weights": true, "license": "Muse Model License" }, "sources": [ { "id": "muse-spark-open-s1", "url": "https://research.meta.ai/blog/muse-spark-open", "type": "official_announcement", "authority": "primary", "archived_url": null, "retrieved": null }, { "id": "muse-spark-open-s2", "url": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B", "type": "official_model_card", "authority": "primary", "archived_url": null, "retrieved": null } ], "provenance": { "status": "partially_verified", "confidence": 60, "reason": "Newly added. Sources are not archived and no value has been traced to one yet — run `npm run enrich`." } } ] }