| Benchmark | Claude 3.5 Sonnet | o1 |
|---|---|---|
| AA Intelligence | 10 | 24 |
| aider_polyglot | 45.3 | 61.7 |
| AIME 2024 | 16 | 79.2 |
| AIME 2025 | 7.4 | 81.7 |
| AIR-Bench 2024 | 85.9 | 80 |
| anthropic_red_team | 99.8 | 98.3 |
| arena_vision | 1146 | 1193 |
| Artificial Analysis Coding Index | 30.2 | 39.7 |
| bbq | 94.9 | 97.3 |
| Codeforces | 717 | 2061 |
| Codeforces (Percentile) | 20.3 | 96.6 |
| Codeforces (Rating) | 717 | 2061 |
| DROP (3-shot F1) | 88.3 | 90.2 |
| Fortress | 13 | 19.4 |
| GPQA Diamond | 67.2 | 75.7 |
| GSM8K | 96.9 | 97.1 |
| harmbench | 98.1 | 96.3 |
| HLE | 3.9 | 7.7 |
| humaneval | 93.7 | 88.1 |
| livecodebench_pass1cot | 36.3 | 63.4 |
| MATH | 81.3 | 96.4 |
| math_500_em | 78.3 | 96.4 |
| MathVista | 67.7 | 71.8 |
| mgsm | 91.6 | 90.8 |
| mmlu | 88.3 | 91.8 |
| MMMU | 72 | 77.6 |
| scicode | 36.6 | 35.8 |
| SEAL VISTA | 38.7 | 45.3 |
| simple_safety_tests | 100 | 99 |
| simplebench | 41.4 | 41.7 |
| simpleqa | 28.4 | 47 |
| SWE-bench Verified | 50.8 | 48.9 |
| TAU-bench (airline) | 46 | 50 |
| TAU-bench (retail) | 69.2 | 70.8 |
| xstest | 95.6 | 97 |
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.