| Benchmark | Claude 3.7 Sonnet | Claude Opus 4.8 |
|---|---|---|
| AA Agentic Index | 37 | 49.4 |
| AA Intelligence | 27.6 | 57.3 |
| AA-LCR | 62.3 | 73 |
| AA-Omniscience | -0.7 | 28.8 |
| arc_agi_1 | 13.6 | 92.5 |
| ARC-AGI-2 | 0 | 72.1 |
| arena_vision | 1195 | 1280 |
| Artificial Analysis Coding Index | 36.4 | 74.3 |
| critpt | 0.9 | 20.9 |
| Fortress | 38 | 18.2 |
| gdpval | 27.4 | 54.2 |
| GPQA Diamond | 84.8 | 93.6 |
| HLE | 9.7 | 57.9 |
| IFBench | 48.3 | 62.2 |
| MMMU-Pro | 60.1 | 78.9 |
| OmniScience Accuracy | 28.2 | 48.8 |
| OmniScience Non-Hallucination | 60 | 60.7 |
| scicode | 40.3 | 53.5 |
| simplebench | 46.4 | 64.8 |
| SWE-bench Verified | 70.3 | 88.6 |
| Terminal-Bench Hard | 21.2 | 58.3 |
| τ²-Bench Telecom (AA run) | 54.7 | 94.4 |
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.