| Benchmark | Claude 3.7 Sonnet | GPT-5 |
|---|---|---|
| AA Agentic Index | 37 | 49.7 |
| AA Intelligence | 27.6 | 35.3 |
| AA-LCR | 62.3 | 76.3 |
| AA-Omniscience | -0.7 | -8.7 |
| aider_polyglot | 64.9 | 88 |
| AIME 2025 | 54.8 | 95 |
| AIR-Bench 2024 | 81.8 | 87.7 |
| arc_agi_1 | 13.6 | 56.2 |
| ARC-AGI-2 | 0 | 7.5 |
| arena_vision | 1195 | 1212 |
| Artificial Analysis Coding Index | 36.4 | 38.9 |
| critpt | 0.9 | 5.7 |
| Fortress | 38 | 17 |
| gdpval | 27.4 | 32.5 |
| GPQA Diamond | 84.8 | 87.3 |
| harmbench | 84.3 | 70 |
| HLE | 9.7 | 28.5 |
| hmmt_feb_2025 | 31.7 | 88.3 |
| IFBench | 48.3 | 73.1 |
| metr_hcast | 54.2 | 68.4 |
| metr_re_bench | 0 | 40 |
| metr_swaa | 98.7 | 100 |
| MMMU | 75 | 84.2 |
| MMMU-Pro | 60.1 | 78.4 |
| multichallenge | 51.6 | 71.1 |
| MultiNRC | 27.8 | 52.1 |
| OmniScience Accuracy | 28.2 | 40.3 |
| OmniScience Non-Hallucination | 60 | 21 |
| scicode | 40.3 | 43 |
| SEAL VISTA | 48.2 | 49.7 |
| simplebench | 46.4 | 56.7 |
| swe_bench_bash | 52.8 | 65 |
| SWE-bench Verified | 70.3 | 74.9 |
| Terminal-Bench | 35.2 | 43.8 |
| Terminal-Bench Hard | 21.2 | 37.9 |
| τ²-Bench Telecom (AA run) | 54.7 | 86.5 |
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.