DeepSeek-R1 ranks first on USAMO 2025.
DeepSeek May 2025 May 2025 snapshot Open weights
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| FaithJudge | 9.78% | #4 of 37 | Official leaderboard | |
| MMLU-Pro | 85% | #27 of 167 | Lab model card | |
| FACTS Grounding | 82.4% | #7 of 34 | Official leaderboard | |
| LongBench v2 | 56.7% | #7 of 33 | Official leaderboard | |
| WritingBench | 78.9 | #10 of 45 | Official leaderboard | |
| Fiction.LiveBench | 75% | #14 of 57 | Fiction.live leaderboard, via Epoch AI | |
| MMLU-Redux | 93.4% | #12 of 45 | Lab model card | |
| Seal-0 | 5.40% | #3 of 11 | Official leaderboard | |
| LMArena (Elo) | 1,421 | #104 of 335 | Official leaderboard | |
| LLM Confabulation Leaderboard | 14.6% | #20 of 58 | Official leaderboard | |
| SimpleQA | 27.8% | #26 of 60 | Lab model card | |
| FRAMES | 83% | #4 of 9 | Lab model card | |
| HUMAINE | 29 | #23 of 50 | Official leaderboard | |
| EQ-Bench 3 | 1,292 | #36 of 78 | Official leaderboard | |
| Creative Writing v3 | 68.3 | #60 of 120 | Official leaderboard | |
| ComplexNLU (Tencent) | 64.5% | #2 of 4 | Official leaderboard | |
| Longform Creative Writing (EQ-Bench) | 59.5 | #64 of 119 | Official leaderboard | |
| ECLeKTic | 17.5% | #19 of 35 | Official leaderboard | |
| MultiNRC | 27.6% | #21 of 38 | Official leaderboard | |
| SimpleQA Verified | 28.6% | #33 of 56 | Official leaderboard | |
| FACTS Parametric | 20.8% | #32 of 45 | Official leaderboard | |
| Vectara Hallucination Leaderboard (2025 dataset) | 11.3% | #74 of 104 | Official leaderboard | |
| Humanity's Last Exam | 17.7% | #115 of 154 | Lab model card | |
| Slop Score | 67.9 | #13 of 17 | Official leaderboard | |
| Vectara Hallucination Leaderboard (HHEM) | 7.70% | #117 of 148 | Official leaderboard | |
| IFEval | 78.4% | #48 of 60 | Official leaderboard | |
| BullshitBench | 12.7% | #92 of 107 | Official leaderboard | |
| Phare | 58.5% | #55 of 62 | Official leaderboard | |
| Global MMLU Lite | 67.4% | #39 of 43 | Official leaderboard | |
| BullshitBench V2 | 8.00% | #118 of 129 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Misguided Attention | 67.1% | #3 of 45 | Official leaderboard | |
| MATH Level 5 | 96.6% | #7 of 94 | Epoch AI run | |
| USAMO 2025 | 30.1% | #1 of 10 | Official leaderboard | |
| AIME 2024 | 91.4% | #14 of 88 | Lab model card | |
| GPQA Main | 66.6% | #15 of 60 | Official leaderboard | |
| SciArena | 1,042 | #9 of 34 | Official leaderboard | |
| CNMO 2024 | 86.9% | #2 of 7 | Lab model card | |
| IneqMath | 9.50% | #10 of 34 | Official leaderboard | |
| PHYBench | 33.5% | #9 of 28 | Official leaderboard | |
| HiPhO | 59.6% | #12 of 36 | Official leaderboard | |
| BRUMO 2025 | 92.5% | #17 of 42 | Official leaderboard | |
| AMO-Bench | 34.3% | #9 of 21 | Official leaderboard | |
| MATH-500 | 92.2% | #17 of 38 | Official leaderboard | |
| AIME 2025 | 87.5% | #64 of 143 | Lab model card | |
| GPQA Diamond | 76.3% | #106 of 229 | Epoch AI run | |
| OTIS Mock AIME 2024-2025 | 66.4% | #119 of 212 | Epoch AI run | |
| SMT 2025 | 83% | #26 of 40 | Official leaderboard | |
| SimpleBench | 40.8% | #73 of 106 | Official leaderboard | |
| HMMT 2025 | 79.4% | #38 of 54 | Lab model card | |
| MathArena (Apex, IMO, USAMO) | 1.04% | #36 of 47 | Official leaderboard | |
| CMIMC 2025 | 71.3% | #27 of 33 | Official leaderboard | |
| ARC-AGI | 21.2% | #80 of 95 | Official leaderboard | |
| ARC-AGI-2 | 1.12% | #83 of 98 | Official leaderboard | |
| IMO-ProofBench | 3.80% | #19 of 20 | Official leaderboard | |
| CORE-Bench | 8.89% | #17 of 17 | Official leaderboard | |
| IMO 2025 | 6.85% | #6 of 6 | Official leaderboard | |
| ATLAS (OpenCompass) | 26.4% | #5 of 5 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveCodeBench | 89.6% | #2 of 21 | Official leaderboard | |
| ArtifactsBench (Gemini 2.5 Pro 0506 judge) | 51.6% | #2 of 18 | Official leaderboard | |
| LiveCodeBench v5 | 74.8% | #3 of 21 | Official leaderboard | |
| AutoCodeBench | 50.2% | #5 of 34 | Official leaderboard | |
| Multi-LCB | 63.1% | #2 of 11 | Official leaderboard | |
| AetherCode | 22.3% | #2 of 9 | Official leaderboard | |
| Aider Polyglot | 71.6% | #9 of 30 | Lab model card | |
| AlgoTune | 1.70× | #7 of 18 | Official leaderboard | |
| METR Time Horizon | 33 min | #14 of 31 | Official leaderboard | |
| LiveSQLBench | 24.5% | #18 of 36 | Official leaderboard | |
| Bird-SQL | 60.9% | #7 of 13 | Official leaderboard | |
| LiveCodeBench v6 | 73.3% | #51 of 90 | Lab model card | |
| MLE-bench | 36.4% | #7 of 12 | Official leaderboard | |
| ArtifactsBench | 47.7% | #12 of 20 | Official leaderboard | |
| DesignArena Data Visualization | 1,180 | #73 of 118 | Official leaderboard | |
| WeirdML | 41.6% | #71 of 112 | Official leaderboard | |
| Codeforces rating | 1,930 | #30 of 47 | Lab model card | |
| DesignArena Website | 1,142 | #102 of 151 | Official leaderboard | |
| DesignArena 3D Design | 1,094 | #96 of 136 | Official leaderboard | |
| DesignArena Game Dev | 1,087 | #103 of 143 | Official leaderboard | |
| DesignArena Code | 1,129 | #104 of 143 | Official leaderboard | |
| DesignArena UI Component | 1,094 | #102 of 139 | Official leaderboard | |
| SWE-bench Verified | 57.6% | #115 of 155 | Lab model card | |
| ALE-Bench | 1,059 | #99 of 126 | Official leaderboard | |
| DesignArena SVG | 1,029 | #84 of 103 | Official leaderboard | |
| Multi-SWE-bench flash | 15.3% | #4 of 4 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Elimination Game | 5 | #8 of 56 | Official leaderboard | |
| Berkeley Function Calling Leaderboard (BFCL) | 63.8% | #19 of 67 | Official leaderboard | |
| Step Race | 3 | #18 of 59 | Official leaderboard | |
| lmgame-Bench | 32.7 | #9 of 23 | Official leaderboard | |
| LLM Chess | 300 | #49 of 123 | Official leaderboard | |
| VitaBench 2.0 | 41.2% | #7 of 17 | Official leaderboard | |
| AgentBench FC | 49.3% | #5 of 12 | Official leaderboard | |
| tau-bench | 58.7% | #13 of 23 | Lab model card | |
| LisanBench | 288.7 | #82 of 133 | Official leaderboard | |
| ITBench | 23.3% | #13 of 13 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LEXam | 55.9% | #7 of 30 | Official leaderboard | |
| MedHELM | 56.5 | #5 of 13 | Official leaderboard | |
| TaxEval v2 | 72.3% | #63 of 129 | Official leaderboard | |
| CorpFin v2 | 54.1% | #89 of 119 | Official leaderboard | |
| PRBench | 36.6% | #43 of 48 | Official leaderboard | |
| LegalBench | 67.3% | #124 of 138 | Official leaderboard | |
| FailureSensorIQ | 29.1% | #24 of 24 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MolecularIQ | 37.1% | #2 of 17 | Official leaderboard | |
| BaxBench | 39% | #10 of 34 | Official leaderboard | |
| Anthropic Red Team | 99.1% | #37 of 75 | Official leaderboard | |
| LiveSecBench | 55.2 | #18 of 34 | Official leaderboard | |
| Lakera B3 (Backbone Breaker Benchmark) | 53.4% | #20 of 35 | Official leaderboard | |
| SimpleSafetyTests | 98.3% | #46 of 75 | Official leaderboard | |
| Human Pathogen Capabilities Test v2 (HPCT v2) | 46.4% | #49 of 71 | Official leaderboard | |
| HELM Safety | 89.4 | #53 of 75 | Official leaderboard | |
| Human Pathogen Capabilities Test | 44.1% | #61 of 86 | Official leaderboard | |
| Molecular Biology Capabilities Test | 43.1% | #64 of 90 | Official leaderboard | |
| AIR-Bench 2024 | 52.9% | #61 of 82 | Official leaderboard | |
| BioTIER Permit | 95.3% | #57 of 69 | Official leaderboard | |
| BioTIER Refuse | 15.6% | #64 of 69 | Official leaderboard | |
| Spiral-Bench v1.2 | 14.2 | #21 of 21 | Official leaderboard | |
| Spiral-Bench v1.0 | 22.4 | #13 of 13 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HELM Capabilities | 69.9 | #22 of 60 | Official leaderboard |
The Self-Reported Index fits one score to the 13 capability results DeepSeek reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 137.2 to 141.1. How we compute it.
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| GPQA Diamond | 81% GPQA-Diamond (Pass@1) | 76.3% Epoch AI run | +4.7 pts |
| HMMT 2025 | 79.4% HMMT 2025 (Pass@1) | 76.7% Leaderboard | +2.7 pts |
| AIME 2025 | 87.5% AIME 2025 (Pass@1) | 89.2% Leaderboard | −1.7 pts |
| MMLU-Pro | 85% MMLU-Pro (EM) | 84% Leaderboard | +1.0 pts |
| LiveCodeBench v6 | 73.3% LiveCodeBench (2408-2505) (Pass@1) | 73.1% Leaderboard | +0.2 pts |
| Aider Polyglot | 71.6% Aider-Polyglot (Acc.) | 71.4% Leaderboard, aider, diff edits | +0.2 pts |
DeepSeek's models in color, against every other model on the Epoch Capabilities Index, by release date.
Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)
The benchmarks the lab chose to headline, with the scores it gave.