DeepSeek Coder V2 — benchmark results
DeepSeek's open MoE code model (236B total/21B active, 128K context, 338 languages) that rivaled GPT-4 Turbo on coding (June 2024). Provider: DeepSeek. Released 2024-06-17. Access: Open.
Unified ELO 1571 ± 17, rank #561 of 1776 rated models, from 28 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LiveBench Math Comp | 50 | Score | 97.2 |
| LiveBench Olympiad | 64.61 | Score | 94.4 |
| LiveBench AMPS Hard | 43 | Score | 93.1 |
| LiveBench Zebra Puzzle | 42 | Score | 91.7 |
| OlympicArena | 29.31 | Overall Accuracy (%) | 88.9 |
| LiveBench Table Join | 30.76 | Score | 87.5 |
| LiveBench Web Of Lies V2 | 56 | Score | 83.3 |
| LLM2014 Logic 2024-06 | 58.35 | Score (%) | 82.8 |
| LiveBench LCB Generation | 40 | Score | 82.6 |
| LiveBench Coding Completion | 42.1 | Score | 81.2 |
| LLM2014 Logic 2024-07 | 58.41 | Score (%) | 80.8 |
| LiveBench Story Generation | 75.33 | Score | 80.6 |
Interactive version: theaggregate.ai/model?slug=deepseek-coder-v2 · How the rankings work · Data refreshed daily, snapshot 2026-07-22.