DeepSeek-V3 Mar 2025
DeepSeek-V3 places in the top three on 2 benchmarks.
89
Benchmarks with a score
0
Ranked first, out of five or more models
2
In the top three
DeepSeek Mar 2025 DeepSeek-V3
Mar 2025 snapshot Open weights
135.2
#141 of 256
Epoch Capabilities Index
- GPQA Diamond 67.6% #122 /229
- SimpleQA Verified 27.4% #35 /56
- AMO-Bench 5.20% #19 /21
Where it ranks
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.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ToRR | 49.7 | #3 of 14 | Official leaderboard | |
| MMLU-Pro | 81.3% | #50 of 212 | Official leaderboard | |
| HUMAINE | 31 | #12 of 50 | Official leaderboard | |
| Vectara Hallucination Leaderboard (2025 dataset) | 6.10% | #26 of 104 | Official leaderboard | |
| ECLeKTic | 22% | #13 of 35 | Official leaderboard | |
| LMArena (Elo) | 1,395 | #136 of 335 | Official leaderboard | |
| AgentIF | 56.7% | #6 of 12 | Official leaderboard | |
| Wer wird Millionär? LLM Benchmark | 3600.0 | #39 of 77 | Official leaderboard | |
| IFEval | 83.2% | #31 of 60 | Official leaderboard | |
| Encyclo-K | 30.9% | #12 of 23 | Official leaderboard | |
| Longform Creative Writing (EQ-Bench) | 60.1 | #63 of 119 | Official leaderboard | |
| Creative Writing v3 | 65.8 | #67 of 120 | Official leaderboard | |
| WritingBench | 69.4 | #26 of 45 | Official leaderboard | |
| MARS-Bench | 37.3% | #9 of 15 | Official leaderboard | |
| SimpleQA Verified | 27.4% | #35 of 56 | Official leaderboard | |
| EQ-Bench 3 | 1,124 | #49 of 78 | Official leaderboard | |
| Phare | 62.2% | #44 of 62 | Official leaderboard | |
| Fiction.LiveBench | 50% | #42 of 57 | Fiction.live leaderboard, via Epoch AI | |
| Vectara Hallucination Leaderboard (HHEM) | 8.00% | #120 of 148 | Official leaderboard | |
| LLM Confabulation Leaderboard | 26.1% | #49 of 58 | Official leaderboard | |
| FACTS Grounding | 63.9% | #29 of 34 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MATH Level 5 | 75.5% | #30 of 94 | Epoch AI run | |
| Misguided Attention | 53.5% | #15 of 45 | Official leaderboard | |
| IneqMath | 7.00% | #13 of 34 | Official leaderboard | |
| GPQA Main | 53.8% | #30 of 60 | Official leaderboard | |
| PolyMath | 30.7% | #20 of 38 | Official leaderboard | |
| GPQA Diamond | 67.6% | #122 of 229 | Epoch AI run | |
| HiPhO | 45.7% | #23 of 36 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 37.8% | #142 of 212 | Epoch AI run | |
| MATH-500 | 88.6% | #26 of 38 | Official leaderboard | |
| DTBench | 64.8% | #126 of 180 | Official leaderboard | |
| CORE-Bench | 17.8% | #13 of 17 | Official leaderboard | |
| SimpleBench | 27.2% | #87 of 106 | Official leaderboard | |
| AIME 2025 | 50% | #45 of 53 | Official leaderboard | |
| PHYBench | 13.6% | #24 of 28 | Official leaderboard | |
| FutureX | 9.14 | #6 of 7 | Official leaderboard | |
| HMMT 2025 | 29.2% | #47 of 53 | Official leaderboard | |
| IMO-ProofBench | 4.30% | #18 of 20 | Official leaderboard | |
| AMO-Bench | 5.20% | #19 of 21 | Official leaderboard | |
| PutnamBench | 0% | #11 of 12 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BigCodeBench-Hard | 31.8% | #13 of 96 | Official leaderboard | |
| AutoCodeBench | 48.1% | #8 of 34 | Official leaderboard | |
| ArtifactsBench (Gemini 2.5 Pro 0506 judge) | 45.6% | #5 of 18 | Official leaderboard | |
| Aider Polyglot | 55.1% | #17 of 46 | Official leaderboard | |
| SWE-bench Lite | 36.7% | #6 of 15 | Official leaderboard | |
| METR Time Horizon | 25 min | #17 of 31 | Official leaderboard | |
| BigO(Bench) | 34.2% | #10 of 18 | Official leaderboard | |
| BIRD-CRITIC | 24.5% | #5 of 9 | Official leaderboard | |
| OIBench | 11.6% | #5 of 9 | Official leaderboard | |
| GitTaskBench | 26.9% | #6 of 10 | Official leaderboard | |
| SWE-bench Verified | 42% | #40 of 60 | Official leaderboard | |
| DesignArena Website | 1,113 | #111 of 151 | Official leaderboard | |
| DesignArena 3D Design | 1,069 | #101 of 136 | Official leaderboard | |
| LiveSQLBench | 19.8% | #27 of 36 | Official leaderboard | |
| DesignArena UI Component | 1,084 | #107 of 139 | Official leaderboard | |
| DesignArena Code | 1,098 | #112 of 143 | Official leaderboard | |
| ArtifactsBench | 43.5% | #16 of 20 | Official leaderboard | |
| DesignArena Game Dev | 1,043 | #118 of 143 | Official leaderboard | |
| AetherCode | 8.50% | #8 of 9 | Official leaderboard | |
| DesignArena Data Visualization | 1,087 | #106 of 118 | Official leaderboard | |
| SWE-bench-Live | 15.3% | #11 of 12 | Official leaderboard | |
| IOI (Vals) | 1.67% | #51 of 55 | Official leaderboard | |
| DesignArena SVG | 967 | #100 of 103 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Berkeley Function Calling Leaderboard (BFCL) | 64.7% | #15 of 67 | Official leaderboard | |
| Galileo Agent Leaderboard | 0.91 | #8 of 29 | Official leaderboard | |
| ToolComp | 64.2% | #9 of 32 | Official leaderboard | |
| Galileo Agent Leaderboard v2 | 40% | #12 of 21 | Official leaderboard | |
| SnakeBench | 21 | #127 of 188 | Official leaderboard | |
| LLM Chess | -579 | #112 of 123 | Official leaderboard | |
| AgentBench FC | 36.1% | #11 of 12 | Official leaderboard | |
| MCP-Universe | 14.3% | #30 of 30 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MedQA | 80.9% | #7 of 71 | Official leaderboard | |
| LEXam | 52.5% | #10 of 30 | Official leaderboard | |
| LegalBench | 80.8% | #80 of 138 | Official leaderboard | |
| TaxEval v2 | 71.1% | #78 of 129 | Official leaderboard | |
| CorpFin v2 | 54.7% | #85 of 119 | Official leaderboard | |
| StockBench | 11.0 | #11 of 13 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BaxBench | 30.5% | #17 of 34 | Official leaderboard | |
| Anthropic Red Team | 97.1% | #58 of 75 | Official leaderboard | |
| HELM Safety | 87.2 | #60 of 75 | Official leaderboard | |
| SimpleSafetyTests | 95.3% | #64 of 75 | Official leaderboard | |
| CyberGym | 3.58% | #30 of 34 | Official leaderboard | |
| AIR-Bench 2024 | 40.8% | #75 of 82 | Official leaderboard | |
| LiveSecBench | 18.1 | #33 of 34 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HELM Lite | 90.8 | #3 of 71 | Official leaderboard | |
| HELM Capabilities | 66.5 | #29 of 60 | Official leaderboard | |
| Conceptual Reasoning Index | 26.2 | #123 of 151 | Official leaderboard | |
| ACCoRD | 43.4 | #126 of 152 | Official leaderboard | |
| LMCA | 15.5% | #124 of 147 | Official leaderboard |
DeepSeek's models over time
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)