| Benchmark | Claude 3.5 Sonnet | Claude 4 Opus |
|---|---|---|
| AA Intelligence | 10 | 31.7 |
| aider_polyglot | 45.3 | 70.7 |
| AIME 2024 | 16 | 76 |
| AIME 2025 | 7.4 | 75.5 |
| AIR-Bench 2024 | 85.9 | 85.7 |
| anthropic_red_team | 99.8 | 98.9 |
| arena_vision | 1146 | 1207 |
| Artificial Analysis Coding Index | 30.2 | 34 |
| bbq | 94.9 | 99.3 |
| CNMO 2024 | 13.1 | 57.6 |
| Fortress | 13 | 27.6 |
| frontiermath_tier_4 | 0 | 4.2 |
| GPQA Diamond | 67.2 | 79.6 |
| harmbench | 98.1 | 91.7 |
| HLE | 3.9 | 12.3 |
| hmmt_feb_2025 | 1.7 | 60 |
| ifeval | 90.1 | 87.4 |
| LiveCodeBench | 32.8 | 70.4 |
| LiveCodeBench v6 | 37.2 | 47.4 |
| longbench_v2 | 41 | 55.6 |
| math_500_em | 78.3 | 98.2 |
| mmlu | 88.3 | 92.9 |
| mmlu_redux | 88.9 | 94.2 |
| MMLU-Pro | 78 | 86.6 |
| MMMU | 72 | 73.7 |
| scicode | 36.6 | 40.9 |
| SEAL VISTA | 38.7 | 47 |
| simple_safety_tests | 100 | 100 |
| simplebench | 41.4 | 58.8 |
| simpleqa | 28.4 | 22.8 |
| supergpqa | 48.2 | 56.5 |
| SWE-bench Verified | 50.8 | 79.4 |
| TAU-bench (airline) | 46 | 59.6 |
| TAU-bench (retail) | 69.2 | 81.4 |
| 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.