| Benchmark | Claude 3.5 Sonnet | Claude Sonnet 4 |
|---|---|---|
| AA Intelligence | 10 | 29.8 |
| aider_polyglot | 45.3 | 70.7 |
| AIME 2024 | 16 | 43.4 |
| AIME 2025 | 7.4 | 74 |
| AIR-Bench 2024 | 85.9 | 88.3 |
| anthropic_red_team | 99.8 | 99.2 |
| arena_vision | 1146 | 1207 |
| Artificial Analysis Coding Index | 30.2 | 37.6 |
| bbq | 94.9 | 97.9 |
| CNMO 2024 | 13.1 | 60.4 |
| Fortress | 13 | 24.4 |
| frontiermath_tier_4 | 0 | 0 |
| GPQA Diamond | 67.2 | 78 |
| harmbench | 98.1 | 98 |
| HLE | 3.9 | 10.7 |
| ifeval | 90.1 | 87.6 |
| LiveCodeBench | 32.8 | 68.5 |
| LiveCodeBench v6 | 37.2 | 48.5 |
| math_500_em | 78.3 | 94 |
| mmlu | 88.3 | 91.5 |
| mmlu_redux | 88.9 | 93.6 |
| MMLU-Pro | 78 | 84 |
| MMMU | 72 | 74.4 |
| MMMU-Pro | 54.7 | 62.4 |
| scicode | 36.6 | 40 |
| SEAL VISTA | 38.7 | 45.5 |
| simple_safety_tests | 100 | 100 |
| simplebench | 41.4 | 45.5 |
| simpleqa | 28.4 | 15.9 |
| supergpqa | 48.2 | 55.7 |
| SWE-bench Verified | 50.8 | 80.2 |
| TAU-bench (airline) | 46 | 60 |
| TAU-bench (retail) | 69.2 | 80.5 |
| xstest | 95.6 | 97.2 |
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.