| Benchmark | Claude Sonnet 4 | GPT-5.6 Sol |
|---|---|---|
| AA Agentic Index | 39.2 | 57.8 |
| AA Intelligence | 29.8 | 61 |
| AA-LCR | 69.7 | 77.7 |
| AA-Omniscience | 0.2 | 22 |
| arc_agi_1 | 63.7 | 97.5 |
| ARC-AGI-2 | 13.6 | 92.5 |
| Artificial Analysis Coding Index | 37.6 | 78.3 |
| browsecomp | 12.2 | 90.4 |
| critpt | 1.1 | 32.3 |
| gdpval | 31.2 | 61.1 |
| GPQA Diamond | 78 | 94.6 |
| HLE | 10.7 | 49.5 |
| HLE (with tools) | 20.3 | 58 |
| IFBench | 55 | 72.7 |
| MMMU-Pro | 62.4 | 84.6 |
| OmniScience Accuracy | 22.7 | 59.4 |
| OmniScience Non-Hallucination | 70.9 | 10.6 |
| OSWorld-Verified | 42.2 | 83 |
| scicode | 40 | 56.9 |
| simplebench | 45.5 | 64.8 |
| SWE-bench Pro | 42.7 | 64.6 |
| TauBench V3 - Banking | 16.7 | 44.3 |
| Terminal-Bench 2.1 | 36.3 | 89.5 |
| Terminal-Bench Hard | 31.1 | 65.9 |
| τ²-Bench Telecom (AA run) | 64.6 | 85.1 |
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.