DeepSeek Coder V2 Instruct: benchmark results
Provider: DeepSeek. Released 2024-06-17.
Aggregate Rating 1595 ± 30, rank #1061 of 10916 rated models, from 9 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| GSM8K | 94.5 | Accuracy (%) | 100 |
| EvalPlus (HumanEval+ & MBPP+) | 78.7 | Pass@1 avg (%) | 93.5 |
| BigCodeBench | 48.2 | Pass@1 (%) | 92 |
| MBPP+ | 75.1 | MBPP+ pass@1 (imported) | 87.5 |
| FullStackBench (2024-12 bilingual) | 58.09 | Pass@1 (%), 3,374 problems: 1,687 English and 1,687 Chinese; | 80.8 |
| EvalPlus | 78.7 | EvalPlus Avg. (imported) | 75 |
| HumanEval+ | 82.3 | HumanEval+ pass@1 (imported) | 65.2 |
| WebApp1K | 70.02 | Pass@1 (%) | 54.5 |
| WebApp1K-React | 0.7 | pass@1 | 14.3 |
Interactive version: theaggregate.ai/model?slug=deepseek-coder-v2-instruct · How It Works · Data refreshed daily, snapshot 2026-11-08.