DeepSeek V2: benchmark results
DeepSeek's 236B open MoE (21B active) with Multi-head Latent Attention and a 128K context. Provider: DeepSeek. Released 2024-05-06. Access: Open.
Unified ELO 1537 ± 1, rank #504 of 1392 rated models, from 9 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ARC Challenge (AI2) | 92.2 | Accuracy (%) | 96.1 |
| WinoGrande | 86.3 | Accuracy (%) | 92.5 |
| HellaSwag | 87.1 | Accuracy (%) | 92.1 |
| Big-Bench Hard | 78.8 | Average (%) | 87.5 |
| PIQA | 83.9 | Accuracy (%) | 86.7 |
| MMLU | 78.4 | Accuracy (%) | 77 |
| MixEval | 51.7 | Score | 68.6 |
| TriviaQA | 80 | Accuracy (%) | 64.1 |
| BenchGecko Score | 86.1 | Composite Score | 42.1 |
Interactive version: theaggregate.ai/model?slug=deepseek-v2 · How It Works · Data refreshed daily, snapshot 2026-09-05.