DeepSeek·released 2024-05-071 source
DeepSeek-V2 (MoE-236B, May 2024) benchmark scores: 7 benchmarks tracked. 0% of its scores are independently reproduced — the rest are self-reported or unverified.
Leads on 0 of 7 benchmarks·best result 89.6 on ARC AI2 (self-reported)·0 of 7 independently reproduced
| Benchmark | Score | Evidence status | Measured |
|---|---|---|---|
| ARC AI2 | 89.6accuracy (%) | self-reported· optimizedT1 | 2024-05-07DeepSeek-V3 Technical Report |
| HellaSwag | 82.8accuracy (%) | self-reported· optimizedT1 | 2024-05-07DeepSeek-V3 Technical Report |
| TriviaQA | 80accuracy (%) | self-reported· optimizedT1 | 2024-05-07DeepSeek-V3 Technical Report |
| Winogrande | 72.6accuracy (%) | self-reported· optimizedT1 | 2024-05-07DeepSeek-V3 Technical Report |
| BBH | 71.73accuracy (%) | self-reported· optimizedT1 | 2024-05-07DeepSeek-V3 Technical Report |
| MMLU | 71.2accuracy (%) | self-reported· optimizedT1 | 2024-05-07DeepSeek-V3 Technical Report |
| PIQA | 67.8accuracy (%) | self-reported· optimizedT1 | 2024-05-07DeepSeek-V3 Technical Report |
Claims drawn from cited facts, not live model generation.
This model was developed by DeepSeek and released in 2024. With 236 billion parameters it is frontier-scale, offering substantial headroom but at correspondingly high computational expense.
3 cited facts
The model is scored on seven benchmarks, but it holds no current top score on any of them. Because these results are self-reported and not yet independently confirmed, they should be treated as claims rather than verified outcomes.
3 cited facts
With an average gap of 9.51 points behind the leader on its benchmarks, this model sits well behind the front of the pack. It does not hold the top score on any benchmark in the leader count, so it has no genuine category leadership yet.
2 cited facts
2 facts cross-checked across data sources: 2 single-source.
Source facts, citations, and refresh stamp for this record.
Sources: curated_param_counts, epoch_benchmarks
DeepSeek-V2 (MoE-236B, May 2024) is an AI model developed by DeepSeek, released 2024-05-07. Its benchmark record below tags every score as independently reproduced, self-reported, or unverified.
DeepSeek-V2 (MoE-236B, May 2024) has recorded scores on 7 benchmarks, each shown with its evidence status.
DeepSeek-V2 (MoE-236B, May 2024) has recorded scores on 7 benchmarks — PIQA, Winogrande, TriviaQA, MMLU, ARC AI2, BBH, and 1 more. The full table above shows each score with its evidence status.
0 of 7 recorded scores (0%) are independently reproduced rather than self-reported by the lab.
DeepSeek-V2 (MoE-236B, May 2024) has 7 tracked claims: 7 self-reported.