| Benchmark | Claude 3.5 Sonnet | Claude Opus 4.5 |
|---|---|---|
| AA Intelligence | 10 | 41.9 |
| AIME 2025 | 7.4 | 92.8 |
| Artificial Analysis Coding Index | 30.2 | 47.8 |
| C-Eval | 76.7 | 92.2 |
| Fortress | 13 | 13.6 |
| frontiermath_tier_4 | 0 | 4.2 |
| GPQA Diamond | 67.2 | 87 |
| HLE | 3.9 | 30.8 |
| hmmt_feb_2025 | 1.7 | 92.9 |
| LiveCodeBench | 32.8 | 87 |
| LiveCodeBench v6 | 37.2 | 84.8 |
| longbench_v2 | 41 | 64.4 |
| MathVista | 67.7 | 80.2 |
| mmlu_redux | 88.9 | 95.6 |
| MMLU-Pro | 78 | 90 |
| MMMU | 72 | 80.7 |
| MMMU-Pro | 54.7 | 74 |
| MMStar | 62.2 | 73.2 |
| OCRBench | 790 | 86.5 |
| RealWorldQA | 60.1 | 77 |
| scicode | 36.6 | 50 |
| SEAL VISTA | 38.7 | 46.4 |
| simplebench | 41.4 | 62 |
| supergpqa | 48.2 | 70.6 |
| SWE-bench Verified | 50.8 | 81.5 |
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.