| Benchmark | Claude 3.7 Sonnet | Claude Sonnet 4.6 |
|---|---|---|
| AA Agentic Index | 37 | 61.6 |
| AA Intelligence | 27.6 | 48.4 |
| AA-LCR | 62.3 | 74 |
| AA-Omniscience | -0.7 | 12.2 |
| arc_agi_1 | 13.6 | 86.5 |
| ARC-AGI-2 | 0 | 60.4 |
| arena_vision | 1195 | 1278 |
| Artificial Analysis Coding Index | 36.4 | 63 |
| critpt | 0.9 | 3.1 |
| gdpval | 27.4 | 54.8 |
| GPQA Diamond | 84.8 | 89.9 |
| harmbench | 84.3 | 24.2 |
| HLE | 9.7 | 49 |
| IFBench | 48.3 | 56.6 |
| mmmlu | 86.1 | 89.3 |
| MMMU-Pro | 60.1 | 75.6 |
| OmniScience Accuracy | 28.2 | 40.9 |
| OmniScience Non-Hallucination | 60 | 51.6 |
| scicode | 40.3 | 47 |
| SWE-bench Verified | 70.3 | 79.6 |
| Terminal-Bench Hard | 21.2 | 59.1 |
| τ²-Bench Telecom (AA run) | 54.7 | 97.9 |
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.