| Benchmark | Claude 3.5 Sonnet | DeepSeek-R1 |
|---|---|---|
| AA Intelligence | 10 | 18.6 |
| aider_polyglot | 45.3 | 71.6 |
| AIME 2024 | 16 | 91.4 |
| AIME 2025 | 7.4 | 87.5 |
| AIR-Bench 2024 | 85.9 | 52.9 |
| AlpacaEval2.0 (LC-winrate) | 52 | 87.6 |
| anthropic_red_team | 99.8 | 99.1 |
| Arena-Hard (GPT-4-1106 judge) | 85.2 | 92.3 |
| ArenaHard (GPT-4-1106) | 85.2 | 92.3 |
| Artificial Analysis Coding Index | 30.2 | 24.6 |
| bbq | 94.9 | 96.6 |
| C-Eval | 76.7 | 91.8 |
| Chinese SimpleQA (C-SimpleQA) | 56.8 | 63.7 |
| CLUEWSC | 85.4 | 92.8 |
| CNMO 2024 | 13.1 | 78.8 |
| Codeforces | 717 | 2029 |
| Codeforces (Percentile) | 20.3 | 96.3 |
| Codeforces (Rating) | 717 | 2029 |
| DROP (3-shot F1) | 88.3 | 92.2 |
| Fortress | 13 | 74.4 |
| FRAMES (Acc.) | 72.5 | 83 |
| GPQA Diamond | 67.2 | 81 |
| harmbench | 98.1 | 54.6 |
| HLE | 3.9 | 17.7 |
| hmmt_feb_2025 | 1.7 | 76.7 |
| ifeval | 90.1 | 83.3 |
| LiveCodeBench | 32.8 | 84.4 |
| livecodebench_pass1cot | 36.3 | 65.9 |
| longbench_v2 | 41 | 58.3 |
| MATH | 81.3 | 97.3 |
| math_500_em | 78.3 | 98 |
| mmlu | 88.3 | 92.9 |
| mmlu_redux | 88.9 | 93.4 |
| MMLU-Pro | 78 | 85 |
| scicode | 36.6 | 35.7 |
| simple_safety_tests | 100 | 98.3 |
| simplebench | 41.4 | 40.8 |
| simpleqa | 28.4 | 92.3 |
| SWE-bench Verified | 50.8 | 57.6 |
| TAU-bench (airline) | 46 | 53.5 |
| TAU-bench (retail) | 69.2 | 63.9 |
| xstest | 95.6 | 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.