gpt-oss-20b
OpenAI · · 159 days after GPT-4.5
Verified
3 of 3 published values traced to a source4 primary sources, 4 archived
Why this status
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”.
What these states mean · How records are judged · Has this record changed?
OpenAI's smaller open-weight model, running in 16GB of memory.
Details
| Released | August 5, 2025stated in openai.com, arxiv.org, huggingface.cogpt-oss-20b was released on August 5, 2025.
|
|---|---|
| Company | OpenAI |
| Family | GPT-OSS |
| Released alongside | gpt-oss-120b |
| Era | Reasoning models |
| Type | Language model · LLM |
| Modalities | Text → Text |
| Capabilities | Agentic, Coding, Function calling, Reasoning, Tool use |
| Weights | Open weights |
| Licence | Apache-2.0 |
| Context window | 131K tokensstated in developers.openai.comgpt-oss-20b supports a context window of 131,072 tokens.
|
| Parameters | 21Bstated in openai.com, huggingface.cogpt-oss-20b has 21,000,000,000 parameters.
|
Current information, elsewhere
This record is a historical one: every figure above is traced to a dated primary source. For what is true today — live pricing, current rankings, availability — these are better maintained than anything here could be.
- Current price and providers OpenRouter ↗
- Current benchmark scores Artificial Analysis ↗
- Weights and model card Hugging Face ↗
Sources
Record status: verified · confidence 90/100 · 4 of 4 primary
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”.
- openai.com Official announcement archived
- developers.openai.com Official documentation archived
- arxiv.org Technical paper archived
- huggingface.co Model card archived
Primary means published by the organisation that made the model. Verified records require at least one.
GPT-OSS family
- gpt-oss-120b August 5, 2025
- gpt-oss-20b August 5, 2025
Compare gpt-oss-20b with another model →
See gpt-oss-20b on the interactive timeline →