| Benchmark | Claude 3.5 Sonnet | o4-mini |
|---|---|---|
| AA Intelligence | 10 | 26.1 |
| aider_polyglot | 45.3 | 72 |
| AIME 2025 | 7.4 | 92.7 |
| AIR-Bench 2024 | 85.9 | 78.5 |
| anthropic_red_team | 99.8 | 98.2 |
| arena_vision | 1146 | 1201 |
| Artificial Analysis Coding Index | 30.2 | 25.6 |
| bbq | 94.9 | 94 |
| Fortress | 13 | 21.5 |
| frontiermath_tier_4 | 0 | 6.3 |
| GPQA Diamond | 67.2 | 81.4 |
| GSM8K | 96.9 | 0 |
| harmbench | 98.1 | 97 |
| HLE | 3.9 | 16.5 |
| LiveCodeBench | 32.8 | 84.5 |
| MathVista | 67.7 | 84.3 |
| MMMU | 72 | 81.6 |
| MMMU-Pro | 54.7 | 69.2 |
| scicode | 36.6 | 46.5 |
| SEAL VISTA | 38.7 | 51.8 |
| simple_safety_tests | 100 | 100 |
| simplebench | 41.4 | 38.7 |
| SWE-bench Verified | 50.8 | 68.1 |
| TAU-bench (airline) | 46 | 49.2 |
| TAU-bench (retail) | 69.2 | 71.8 |
| xstest | 95.6 | 97.4 |
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.