DeepSeek-V3
DeepSeek · · 112 days after DeepSeek-V2.5
Verified
3 of 3 published values traced to a source4 primary sources, 4 archived
Why this status
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.
What these states mean · How records are judged · Has this record changed?
A 671B MoE open model whose near-frontier performance at minimal cost stunned the industry.
Details
| Released | December 26, 2024stated in huggingface.co, api-docs.deepseek.com, api-docs.deepseek.comDeepSeek-V3 was released on December 26, 2024.
|
|---|---|
| Company | DeepSeek |
| Family | DeepSeek |
| Lineage | after DeepSeek-V2.5 · before DeepSeek-V3.1 |
| Era | Multimodal & open weights |
| Type | Language model · LLM |
| Modalities | Text → Text |
| Capabilities | Not recorded |
| Weights | Open weights |
| Licence | DeepSeek License Agreement 1.0 |
| Context window | 128K tokensstated in huggingface.coDeepSeek-V3 supports a context window of 128,000 tokens.
|
| Parameters | 671Bstated in arxiv.org, huggingface.co, api-docs.deepseek.comDeepSeek-V3 has 671,000,000,000 parameters.
|
Pricing
The price a lab's own page showed on the day that page was captured. That is what a snapshot can prove; it is not necessarily the launch price, and prices are often cut later. Citing the live pricing page instead would give a number with no date at all.
| Input | $0.27 per million tokens observed March 6, 2025 · archived snapshot (2025-03-06) |
|---|---|
| Output | $1.1 per million tokens observed March 6, 2025 · archived snapshot (2025-03-06) |
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
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.
- arxiv.org Technical paper archived
- huggingface.co Model card archived
- api-docs.deepseek.com Official documentation archived
- api-docs.deepseek.com Official documentation archived
Primary means published by the organisation that made the model. Verified records require at least one.
DeepSeek family
- DeepSeek-V2 May 6, 2024
- DeepSeek-V2.5 September 5, 2024
- DeepSeek-V3 December 26, 2024
- DeepSeek-R1 January 20, 2025
- DeepSeek-V3.1 August 21, 2025
- DeepSeek V4 Flash April 24, 2026
- DeepSeek-V4.1-Flash September 10, 2026
What changed from DeepSeek-V2.5
Compared with the previous release in this series. A field appears only when both records state a value — where one does not, it is listed as uncomparable rather than dropped, because a gap in our research is not a finding about the model.
- Parameters236B → 671B
Not comparable: Context window (not disclosed for both releases)
Compare DeepSeek-V3 with another model →
See DeepSeek-V3 on the interactive timeline →