| Benchmark | Claude 3.5 Sonnet | Claude Sonnet 4.5 |
|---|---|---|
| AA Intelligence | 10 | 37.4 |
| AI2D | 94.7 | 87 |
| AIME 2025 | 7.4 | 88 |
| AIR-Bench 2024 | 85.9 | 89.8 |
| Artificial Analysis Coding Index | 30.2 | 52.1 |
| Fortress | 13 | 17 |
| frontiermath_tier_4 | 0 | 4.2 |
| GPQA Diamond | 67.2 | 83.4 |
| HLE | 3.9 | 19.8 |
| hmmt_feb_2025 | 1.7 | 79.2 |
| LiveCodeBench | 32.8 | 71 |
| LiveCodeBench v6 | 37.2 | 64 |
| longbench_v2 | 41 | 61.8 |
| MathVista | 67.7 | 79.8 |
| mmlu_redux | 88.9 | 95.6 |
| MMLU-Pro | 78 | 88.2 |
| MMMU | 72 | 79.6 |
| MMMU (val) (Pass@1) | 68.3 | 77.8 |
| MMMU-Pro | 54.7 | 68.7 |
| RealWorldQA | 60.1 | 70.3 |
| scicode | 36.6 | 45 |
| SEAL VISTA | 38.7 | 48.8 |
| simplebench | 41.4 | 54.3 |
| SWE-bench Verified | 50.8 | 82 |
| TAU-bench (airline) | 46 | 70 |
| TAU-bench (retail) | 69.2 | 86.2 |
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.