| Benchmark | Claude 3.7 Sonnet | GPT-4o |
|---|---|---|
| AA Agentic Index | 37 | 9.7 |
| AA Intelligence | 27.6 | 11.2 |
| AA-LCR | 62.3 | 55 |
| AA-Omniscience | -0.7 | -10.5 |
| aider_polyglot | 64.9 | 30.7 |
| AIME 2025 | 54.8 | 11.7 |
| AIR-Bench 2024 | 81.8 | 62.4 |
| anthropic_red_team | 99.7 | 99.1 |
| arc_agi_1 | 13.6 | 4.5 |
| ARC-AGI-2 | 0 | 0 |
| arena_vision | 1195 | 1162 |
| Artificial Analysis Coding Index | 36.4 | 24.2 |
| bbq | 92.1 | 95.1 |
| critpt | 0.9 | 0 |
| Fortress | 38 | 47.2 |
| gdpval | 27.4 | 0 |
| GPQA Diamond | 84.8 | 70.1 |
| harmbench | 84.3 | 82.9 |
| HLE | 9.7 | 5.3 |
| IFBench | 48.3 | 36 |
| ifeval | 93.2 | 84.3 |
| math_500_em | 96.2 | 74.6 |
| metr_hcast | 54.2 | 25 |
| metr_re_bench | 0 | 0 |
| metr_swaa | 98.7 | 99.2 |
| mmmlu | 86.1 | 81.4 |
| MMMU | 75 | 72.2 |
| MMMU-Pro | 60.1 | 59.9 |
| multichallenge | 51.6 | 40.3 |
| MultiNRC | 27.8 | 12.4 |
| OmniScience Accuracy | 28.2 | 23.7 |
| OmniScience Non-Hallucination | 60 | 62.1 |
| scicode | 40.3 | 33.4 |
| SEAL VISTA | 48.2 | 34.9 |
| simple_safety_tests | 100 | 98.5 |
| simplebench | 46.4 | 17.8 |
| swe_bench_bash | 52.8 | 21.6 |
| SWE-bench Verified | 70.3 | 38.8 |
| TAU-bench (airline) | 58.4 | 42.8 |
| TAU-bench (retail) | 81.2 | 60.3 |
| Terminal-Bench Hard | 21.2 | 8.3 |
| xstest | 96.4 | 97.3 |
| τ²-Bench Telecom (AA run) | 54.7 | 28.9 |
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.