| Benchmark | Claude 3.7 Sonnet | Claude Sonnet 4.5 |
|---|---|---|
| AA Agentic Index | 37 | 50.6 |
| AA Intelligence | 27.6 | 37.4 |
| AA-LCR | 62.3 | 68.3 |
| AA-Omniscience | -0.7 | -0.1 |
| AIME 2025 | 54.8 | 88 |
| AIR-Bench 2024 | 81.8 | 89.8 |
| arc_agi_1 | 13.6 | 63.7 |
| ARC-AGI-2 | 0 | 13.6 |
| Artificial Analysis Coding Index | 36.4 | 52.1 |
| critpt | 0.9 | 1.1 |
| Fortress | 38 | 17 |
| gdpval | 27.4 | 40.4 |
| GPQA Diamond | 84.8 | 83.4 |
| HLE | 9.7 | 19.8 |
| hmmt_feb_2025 | 31.7 | 79.2 |
| IFBench | 48.3 | 57.3 |
| mmmlu | 86.1 | 89.1 |
| MMMU | 75 | 79.6 |
| MMMU-Pro | 60.1 | 68.7 |
| multichallenge | 51.6 | 55.3 |
| MultiNRC | 27.8 | 35.8 |
| OmniScience Accuracy | 28.2 | 32.9 |
| OmniScience Non-Hallucination | 60 | 50.9 |
| scicode | 40.3 | 45 |
| SEAL VISTA | 48.2 | 48.8 |
| simplebench | 46.4 | 54.3 |
| swe_bench_bash | 52.8 | 71.4 |
| SWE-bench Verified | 70.3 | 82 |
| TAU-bench (airline) | 58.4 | 70 |
| TAU-bench (retail) | 81.2 | 86.2 |
| Terminal-Bench | 35.2 | 51 |
| Terminal-Bench Hard | 21.2 | 35.6 |
| τ²-Bench Telecom (AA run) | 54.7 | 78.1 |
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.