DeepSeek·released 2025-03-241 source
DeepSeek-V3 (Mar 2025) benchmark scores: 11 benchmarks tracked. 27% of its scores are independently reproduced — the rest are self-reported or unverified.
Leads on 0 of 11 benchmarks·best result 77 on Lech Mazur Writing (unverified)·3 of 11 independently reproduced
| Benchmark | Score | Evidence status | Measured |
|---|---|---|---|
| Lech Mazur Writing | 77accuracy (%) | unverifiedT1 | 2025-03-24lechmazur/writing Github repository |
| MATH level 5 | 75.55accuracy (%) | reproduced· optimizedT1 | 2025-03-24Epoch AI |
| GPQA diamond | 56.82accuracy (%) | reproduced· optimizedT1 | 2025-03-24Epoch AI |
| Aider polyglot | 55.1accuracy (%) | unverified· optimizedT1 | 2025-03-24https://aider.chat/docs/leaderboards/ |
| Fiction.LiveBench | 50accuracy (%) | unverifiedT1 | 2025-03-24Fiction.live leaderboard |
| METR Time Horizons | 49.58accuracy (%) | unverifiedT1 | 2025-03-24METR - Measuring AI Ability to Complete Long Tasks |
| DTBench | 41.33accuracy (%) | unverifiedT1 | 2025-03-24https://conceptualreasoning.ai/dtbench |
| OTIS Mock AIME 2024-2025 | 37.72accuracy (%) | reproduced· optimizedT1 | 2025-03-24Epoch AI |
| WeirdML | 36.08accuracy (%) | unverifiedT1 | 2025-03-24https://htihle.github.io/weirdml.html |
| LMCA | 18.2accuracy (%) | unverifiedT1 | 2025-03-24https://conceptualreasoning.ai/lmca |
| SimpleBench | 12.64accuracy (%) | unverifiedT1 | 2025-03-24SimpleBench Leaderboard |
Claims drawn from cited facts, not live model generation.
This model was developed by DeepSeek and released in 2025.
2 cited facts
This model is scored on 11 tracked benchmarks, and it currently holds no top score on any of them. Its results have been independently reproduced, meaning outside evaluation backs up the reported scores.
3 cited facts
Against the best reported scores on its benchmarks, this model trails the leader by an average of 43.23 points, a gap wide enough to read as well behind the front of the pack rather than merely competitive. Consistent with that margin, it holds the top score on none of the benchmarks it reports, meaning there is no category where it currently stands as genuine state-of-the-art leadership.
2 cited facts
1 facts cross-checked across data sources: 1 single-source.
Source facts, citations, and refresh stamp for this record.
Sources: epoch_benchmarks
DeepSeek-V3 (Mar 2025) is an AI model developed by DeepSeek, released 2025-03-24. Its benchmark record below tags every score as independently reproduced, self-reported, or unverified.
DeepSeek-V3 (Mar 2025) has recorded scores on 11 benchmarks, each shown with its evidence status.
DeepSeek-V3 (Mar 2025) has recorded scores on 11 benchmarks — Lech Mazur Writing, GPQA diamond, MATH level 5, OTIS Mock AIME 2024-2025, DTBench, WeirdML, and 5 more. The full table above shows each score with its evidence status.
3 of 11 recorded scores (27%) are independently reproduced rather than self-reported by the lab.
DeepSeek-V3 (Mar 2025) has 11 tracked claims: 3 independently reproduced, 8 unverified.