| Benchmark | Claude Opus 4.7 | o3 |
|---|---|---|
| AA Agentic Index | 64.6 | 36.1 |
| AA Intelligence | 55 | 31.1 |
| AA-LCR | 75.3 | 73.3 |
| AA-Omniscience | 27.3 | -15.3 |
| arc_agi_1 | 92 | 60.8 |
| ARC-AGI-2 | 75.8 | 6.5 |
| arena_vision | 1304 | 1217 |
| Artificial Analysis Coding Index | 73.6 | 38.4 |
| browsecomp | 79.8 | 49.7 |
| critpt | 12 | 1.1 |
| frontiermath_tier_4 | 22.9 | 2.1 |
| gdpval | 58.6 | 12.8 |
| GPQA Diamond | 94.2 | 83.3 |
| harmbench | 17.5 | 98.4 |
| HLE | 54.7 | 20.3 |
| IFBench | 58.6 | 71.4 |
| MMMU-Pro | 78.8 | 76.4 |
| OmniScience Accuracy | 48.9 | 38.6 |
| OmniScience Non-Hallucination | 57.7 | 12.9 |
| scicode | 54.5 | 41 |
| simplebench | 61.7 | 53.1 |
| simpleqa_verified | 50.6 | 53 |
| SWE-bench Verified | 87.6 | 69.1 |
| Terminal-Bench Hard | 54.5 | 37.1 |
| τ²-Bench Telecom (AA run) | 88.6 | 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.