Updated daily at 05:00 UTC
- Released
- 26 Dec 2024
- Status
- Legacy
- Context
- 131.1K
- Max output
- —
- Input per 1M tokens
- —
- Output per 1M tokens
- —
- Open weights
- Yes
- License
- mit
- Parameters
- 671B, 37B active
- Knowledge cutoff
- —
- Reasoning
- No
- Modalities
- text → text
Category scores
Out of 100, from independent and third-party results only, by method category-2026-09-29.1.
| Category | Score | Rank | Coverage |
|---|---|---|---|
| Knowledge & reasoning | 2.7 | Not ranked | 1 of 4 |
| Coding & software engineering | 48.4 | Not ranked | 1 of 5 |
| Human preference | 44.1 | Not ranked | 1 of 5 |
Benchmark results
Every result we hold for this model, with who ran it and where it was published.
Knowledge & reasoning
| Benchmark | Metric | Value | Provenance | Setting | Evaluated | Source |
|---|---|---|---|---|---|---|
| AI2 Reasoning Challenge (ARC) | Challenge set accuracy | 95.3% | Self-reported | — | — | DeepSeek-V3 Technical Report via Epoch AI |
| BIG-Bench Hard (BBH) | Average accuracy | 87.5% | Self-reported | — | — | DeepSeek-V3 Technical Report via Epoch AI |
| BIG-Bench Hard (BBH) | Average accuracy | 82.9% | Self-reported | — | — | DeepSeek-V3 Technical Report via Epoch AI |
| HellaSwag | Accuracy | 88.9% | Self-reported | — | — | DeepSeek-V3 Technical Report via Epoch AI |
| MMLU | Accuracy | 87.2% | Third party | — | — | Stanford CRFM Leaderboard via Epoch AI |
| MMLU | Accuracy | 87.1% | Self-reported | — | — | DeepSeek-V3 Technical Report via Epoch AI |
| MMLU-Pro | Accuracy | 64.4% | Self-reported | — | — | Model Card via Hugging Face Community Evals |
| SimpleBench | Score (AVG@5) | 18.9% | Third party | — | — | SimpleBench Leaderboard via Epoch AI |
| WinoGrande | Accuracy | 85.2% | Self-reported | — | — | DeepSeek-V3 Technical Report via Epoch AI |
Scientific reasoning
| Benchmark | Metric | Value | Provenance | Setting | Evaluated | Source |
|---|---|---|---|---|---|---|
| GPQA Diamond | Accuracy | 56.5% | Independent | — | 27 Jan 2025 | Epoch AI |
| GPQA Diamond | Accuracy | 58.2% | Self-reported | — | — | EvalEval via Hugging Face Community Evals |
Mathematics
| Benchmark | Metric | Value | Provenance | Setting | Evaluated | Source |
|---|---|---|---|---|---|---|
| FrontierMath Tiers 1-3 (v1) | Accuracy | 1.7% | Independent | — | 7 Mar 2025 | Epoch AI |
| MATH Level 5 | Accuracy | 64.9% | Independent | — | 27 Jan 2025 | Epoch AI |
| OTIS Mock AIME 2024-2025 | Accuracy | 15.8% | Independent | — | 25 Feb 2025 | Epoch AI |
Coding & software engineering
| Benchmark | Metric | Value | Provenance | Setting | Evaluated | Source |
|---|---|---|---|---|---|---|
| Aider Polyglot | Percent correct | 48.4% | Third party | — | — | Aider LLM Leaderboards via Epoch AI |
| METR Time Horizons | Average task success (Epoch's column) | 47.4% | Third party | — | — | METR - Measuring AI Ability to Complete Long Tasks via Epoch AI |
Factuality & hallucination
| Benchmark | Metric | Value | Provenance | Setting | Evaluated | Source |
|---|---|---|---|---|---|---|
| TriviaQA | Exact match | 82.9% | Self-reported | — | — | DeepSeek-V3 Technical Report via Epoch AI |
Human preference
| Benchmark | Metric | Value | Provenance | Setting | Evaluated | Source |
|---|---|---|---|---|---|---|
| Text Arena (arena.ai) | Arena score (Elo scale) | 1358 | Independent | — | 25 Sep 2026 | Arena |
| Text Arena (arena.ai) | Votes | 21770 | Independent | — | 25 Sep 2026 | Arena |
Other DeepSeek-V versions
- DeepSeek-V4.1-Flash10 Sep 2026
- DeepSeek-V4-Flash-Vision-Exp21 Aug 2026
- DeepSeek-V4-Flash24 Apr 2026
- DeepSeek-V4-Pro24 Apr 2026
- DeepSeek-V3.21 Dec 2025
- DeepSeek-V3.2-Speciale1 Dec 2025
- DeepSeek-V3.2-Exp29 Sep 2025
- DeepSeek-V3.1-Terminus22 Sep 2025
- DeepSeek-V3.121 Aug 2025
- DeepSeek-V2.55 Sep 2024
- DeepSeek-V2-Lite16 May 2024
- DeepSeek-V26 May 2024