| Benchmark | Claude 3.5 Sonnet | DeepSeek-V4-Pro |
|---|---|---|
| AA Intelligence | 10 | 53 |
| Artificial Analysis Coding Index | 30.2 | 59.4 |
| Chinese SimpleQA (C-SimpleQA) | 56.8 | 77.7 |
| Codeforces | 717 | 3206 |
| GPQA Diamond | 67.2 | 90.5 |
| HLE | 3.9 | 48.2 |
| LiveCodeBench | 32.8 | 93.5 |
| MMLU-Pro | 78 | 87.5 |
| scicode | 36.6 | 50 |
| simplebench | 41.4 | 50.9 |
| simpleqa | 28.4 | 57.9 |
| SWE-bench Verified | 50.8 | 80.6 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.