| Benchmark | DeepSeek-R1 | Gemini 1.5 Pro |
|---|---|---|
| AA Intelligence | 18.6 | 10 |
| aider_polyglot | 71.6 | 16.9 |
| AIR-Bench 2024 | 52.9 | 82.8 |
| anthropic_red_team | 99.1 | 99.9 |
| ARC-AGI-2 | 1.3 | 0.8 |
| Artificial Analysis Coding Index | 24.6 | 23.6 |
| bbq | 96.6 | 94.5 |
| Chinese SimpleQA (C-SimpleQA) | 63.7 | 59.4 |
| Fortress | 74.4 | 53.9 |
| GPQA Diamond | 81 | 59.1 |
| harmbench | 54.6 | 79.9 |
| HLE | 17.7 | 4.9 |
| ifeval | 83.3 | 89.4 |
| LiveCodeBench | 84.4 | 30.5 |
| MATH | 97.3 | 92 |
| mmlu | 92.9 | 86.8 |
| MMLU-Pro | 85 | 75.8 |
| scicode | 35.7 | 29.5 |
| simple_safety_tests | 98.3 | 97.5 |
| simplebench | 40.8 | 27.1 |
| simpleqa | 92.3 | 23.4 |
| SWE-bench Verified | 57.6 | 34.2 |
| xstest | 98.8 | 98.8 |
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.