| Benchmark | Claude 3.7 Sonnet | o4-mini |
|---|---|---|
| AA Agentic Index | 37 | 36.1 |
| AA Intelligence | 27.6 | 26.1 |
| AA-LCR | 62.3 | 60 |
| AA-Omniscience | -0.7 | -35.7 |
| aider_polyglot | 64.9 | 72 |
| AIME 2025 | 54.8 | 92.7 |
| AIR-Bench 2024 | 81.8 | 78.5 |
| anthropic_red_team | 99.7 | 98.2 |
| arc_agi_1 | 13.6 | 58.7 |
| ARC-AGI-2 | 0 | 6.1 |
| arena_vision | 1195 | 1201 |
| Artificial Analysis Coding Index | 36.4 | 25.6 |
| bbq | 92.1 | 94 |
| critpt | 0.9 | 0.6 |
| Fortress | 38 | 21.5 |
| gdpval | 27.4 | 25.4 |
| GPQA Diamond | 84.8 | 81.4 |
| harmbench | 84.3 | 97 |
| HLE | 9.7 | 16.5 |
| IFBench | 48.3 | 68.7 |
| MMMU | 75 | 81.6 |
| MMMU-Pro | 60.1 | 69.2 |
| multichallenge | 51.6 | 44.9 |
| MultiNRC | 27.8 | 22.2 |
| OmniScience Accuracy | 28.2 | 24.8 |
| OmniScience Non-Hallucination | 60 | 19.5 |
| scicode | 40.3 | 46.5 |
| SEAL VISTA | 48.2 | 51.8 |
| simple_safety_tests | 100 | 100 |
| simplebench | 46.4 | 38.7 |
| swe_bench_bash | 52.8 | 45 |
| SWE-bench Verified | 70.3 | 68.1 |
| TAU-bench (airline) | 58.4 | 49.2 |
| TAU-bench (retail) | 81.2 | 71.8 |
| Terminal-Bench Hard | 21.2 | 15.2 |
| usamo_2025 | 3.6 | 19.1 |
| xstest | 96.4 | 97.4 |
| τ²-Bench Telecom (AA run) | 54.7 | 55.6 |
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.