| Benchmark | Claude 3.7 Sonnet | Claude Opus 4.5 |
|---|---|---|
| AA Agentic Index | 37 | 59.6 |
| AA Intelligence | 27.6 | 41.9 |
| AA-LCR | 62.3 | 76 |
| AA-Omniscience | -0.7 | 14 |
| AIME 2025 | 54.8 | 92.8 |
| arc_agi_1 | 13.6 | 82 |
| ARC-AGI-2 | 0 | 37.6 |
| Artificial Analysis Coding Index | 36.4 | 47.8 |
| critpt | 0.9 | 4.6 |
| Fortress | 38 | 13.6 |
| gdpval | 27.4 | 47.3 |
| GPQA Diamond | 84.8 | 87 |
| HLE | 9.7 | 30.8 |
| hmmt_feb_2025 | 31.7 | 92.9 |
| IFBench | 48.3 | 58 |
| metr_hcast | 54.2 | 73.4 |
| metr_re_bench | 0 | 40 |
| metr_swaa | 98.7 | 99.5 |
| mmmlu | 86.1 | 90.8 |
| MMMU | 75 | 80.7 |
| MMMU-Pro | 60.1 | 74 |
| multichallenge | 51.6 | 59 |
| MultiNRC | 27.8 | 48.6 |
| OmniScience Accuracy | 28.2 | 46.6 |
| OmniScience Non-Hallucination | 60 | 39 |
| scicode | 40.3 | 50 |
| SEAL VISTA | 48.2 | 46.4 |
| simplebench | 46.4 | 62 |
| swe_bench_bash | 52.8 | 76.8 |
| SWE-bench Verified | 70.3 | 81.5 |
| Terminal-Bench Hard | 21.2 | 47 |
| τ²-Bench Telecom (AA run) | 54.7 | 89.5 |
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.