| Benchmark | Claude Sonnet 4.6 | o3 |
|---|---|---|
| AA Agentic Index | 61.6 | 36.1 |
| AA Intelligence | 48.4 | 31.1 |
| AA-LCR | 74 | 73.3 |
| AA-Omniscience | 12.2 | -15.3 |
| arc_agi_1 | 86.5 | 60.8 |
| ARC-AGI-2 | 60.4 | 6.5 |
| arena_vision | 1278 | 1217 |
| Artificial Analysis Coding Index | 63 | 38.4 |
| browsecomp | 76.2 | 49.7 |
| critpt | 3.1 | 1.1 |
| frontiermath_tier_4 | 8.3 | 2.1 |
| gdpval | 54.8 | 12.8 |
| GPQA Diamond | 89.9 | 83.3 |
| harmbench | 24.2 | 98.4 |
| HLE | 49 | 20.3 |
| IFBench | 56.6 | 71.4 |
| MMMU-Pro | 75.6 | 76.4 |
| OmniScience Accuracy | 40.9 | 38.6 |
| OmniScience Non-Hallucination | 51.6 | 12.9 |
| scicode | 47 | 41 |
| simpleqa_verified | 29 | 53 |
| SWE-bench Verified | 79.6 | 69.1 |
| Tau2 retail | 91.7 | 80.2 |
| Terminal-Bench Hard | 59.1 | 37.1 |
| τ²-Bench Telecom (AA run) | 97.9 | 80.7 |
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.