DeepSeek-V3.2-Exp ranks first on 2 of the 49 benchmarks with five or more models scored.
DeepSeek Sep 2025 Open weights Flagship
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 | |
|---|---|---|---|---|
| SimpleQA | 97.1% | #1 of 60 | Lab model card | |
| Vectara Hallucination Leaderboard (2025 dataset) | 5.30% | #15 of 104 | Official leaderboard | |
| MMLU-Pro | 85% | #29 of 167 | Lab model card | |
| Fiction.LiveBench | 83.3% | #11 of 57 | Fiction.live leaderboard, via Epoch AI | |
| LMArena (Elo) | 1,425 | #97 of 335 | Official leaderboard | |
| ConStory-Bench | 0.54 | #6 of 14 | Official leaderboard | |
| Longform Creative Writing (EQ-Bench) | 60.2 | #62 of 119 | Official leaderboard | |
| Slop Score | 54.2 | #9 of 17 | Official leaderboard | |
| Vectara Hallucination Leaderboard (HHEM) | 4.60% | #85 of 148 | Official leaderboard | |
| Humanity's Last Exam | 19.8% | #105 of 154 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BRUMO 2025 | 95.8% | #12 of 42 | Official leaderboard | |
| DTBench | 87.7% | #61 of 180 | Official leaderboard | |
| AIME 2025 | 89.3% | #53 of 143 | Lab model card | |
| GPQA Diamond | 79.9% | #119 of 302 | Lab model card | |
| SMT 2025 | 84.9% | #21 of 40 | Official leaderboard | |
| CMIMC 2025 | 76.3% | #21 of 33 | Official leaderboard | |
| HMMT 2025 | 83.6% | #35 of 54 | Lab model card | |
| HMMT Nov 2025 | 84.2% | #18 of 22 | Official leaderboard | |
| MathArena (Apex, IMO, USAMO) | 0.52% | #44 of 47 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BigCodeArena | 1,117 | #1 of 13 | Official leaderboard | |
| Aider Polyglot | 74.5% | #7 of 30 | Lab model card | |
| LiveCodeBench | 74.1% | #14 of 54 | Lab model card | |
| Codeforces rating | 2,121 | #17 of 47 | Lab model card | |
| LiveCodeBench v6 | 74.1% | #49 of 90 | Lab model card | |
| DesignArena 3D Design | 1,131 | #80 of 136 | Official leaderboard | |
| DesignArena Website | 1,172 | #89 of 151 | Official leaderboard | |
| DesignArena UI Component | 1,158 | #84 of 139 | Official leaderboard | |
| SWE-bench Verified | 67.8% | #94 of 155 | Lab model card | |
| DesignArena Code | 1,161 | #88 of 143 | Official leaderboard | |
| DesignArena Game Dev | 1,134 | #90 of 143 | Official leaderboard | |
| WeirdML | 39.5% | #79 of 112 | Official leaderboard | |
| WebDev Arena | 1,367 | #77 of 105 | WebDev Arena Leaderboard, via Epoch AI | |
| DesignArena Data Visualization | 1,150 | #91 of 118 | Official leaderboard | |
| SWE-bench Multilingual | 57.9% | #61 of 73 | Lab model card | |
| DesignArena SVG | 1,022 | #87 of 103 | Official leaderboard | |
| Code Arena | 1,270 | #112 of 128 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BFCL V4 | 56.7% | #11 of 65 | Official leaderboard | |
| Step Race | 4 | #10 of 59 | Official leaderboard | |
| Terminal-Bench | 37.7% | #13 of 22 | Lab model card | |
| Toolathlon | 20.1% | #38 of 50 | Official leaderboard | |
| MCP-Universe | 19.9% | #25 of 30 | Official leaderboard | |
| BrowseComp | 40.1% | #71 of 81 | Lab model card | |
| BrowseComp-ZH | 47.9% | #18 of 19 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| nocot-bench (NCRI) | 76.7 | #69 of 185 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Conceptual Reasoning Index | 45.7 | #84 of 151 | Official leaderboard | |
| ACCoRD | 61.8 | #85 of 152 | Official leaderboard | |
| LMCA | 29.1% | #95 of 147 | Official leaderboard |
The Self-Reported Index fits one score to the 14 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 140.0 to 147.4. 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.
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.
Every result in the longer documents the lab published with it. 1 of them isn't in the launch post.