| Benchmark | Claude Sonnet 4.5 | o3 |
|---|---|---|
| AA Agentic Index | 50.6 | 36.1 |
| AA Intelligence | 37.4 | 31.1 |
| AA-LCR | 68.3 | 73.3 |
| AA-Omniscience | -0.1 | -15.3 |
| AIME 2025 | 88 | 89.2 |
| AIR-Bench 2024 | 89.8 | 84.5 |
| arc_agi_1 | 63.7 | 60.8 |
| ARC-AGI-2 | 13.6 | 6.5 |
| Artificial Analysis Coding Index | 52.1 | 38.4 |
| browsecomp | 24.1 | 49.7 |
| critpt | 1.1 | 1.1 |
| Fortress | 17 | 16 |
| frontiermath_tier_4 | 4.2 | 2.1 |
| gdpval | 40.4 | 12.8 |
| GPQA Diamond | 83.4 | 83.3 |
| HLE | 19.8 | 20.3 |
| hmmt_feb_2025 | 79.2 | 77.5 |
| IFBench | 57.3 | 71.4 |
| LiveCodeBench | 71 | 84.7 |
| longbench_v2 | 61.8 | 58.8 |
| MathVista | 79.8 | 86.8 |
| MMLU-Pro | 88.2 | 85 |
| MMMU | 79.6 | 82.9 |
| MMMU-Pro | 68.7 | 76.4 |
| multichallenge | 55.3 | 60.4 |
| MultiNRC | 35.8 | 45.5 |
| OmniScience Accuracy | 32.9 | 38.6 |
| OmniScience Non-Hallucination | 50.9 | 12.9 |
| scicode | 45 | 41 |
| simplebench | 54.3 | 53.1 |
| simpleqa_verified | 23.6 | 53 |
| swe_bench_bash | 71.4 | 58.4 |
| SWE-bench Verified | 82 | 69.1 |
| TAU-bench (airline) | 70 | 52 |
| TAU-bench (retail) | 86.2 | 73.9 |
| Terminal-Bench Hard | 35.6 | 37.1 |
| VisualToolBench | 6.2 | 13.7 |
| τ²-Bench Telecom (AA run) | 78.1 | 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.