| Benchmark | Claude Sonnet 4.5 | GPT-4o mini |
|---|---|---|
| AA Agentic Index | 50.6 | 1 |
| AA Intelligence | 37.4 | 7 |
| AI2D | 87 | 75.2 |
| AIR-Bench 2024 | 89.8 | 56.3 |
| ARC-AGI-2 | 13.6 | 0 |
| Artificial Analysis Coding Index | 52.1 | 11.4 |
| Fortress | 17 | 48.1 |
| gdpval | 40.4 | 0 |
| GPQA Diamond | 83.4 | 42.6 |
| HLE | 19.8 | 4.2 |
| IFBench | 57.3 | 31 |
| LiveCodeBench | 71 | 35.5 |
| MathVista | 79.8 | 56.7 |
| MMMU | 79.6 | 59.4 |
| MMMU (val) (Pass@1) | 77.8 | 52.1 |
| MMMU-Pro | 68.7 | 41.5 |
| scicode | 45 | 22.9 |
| simplebench | 54.3 | 10.7 |
| SWE-bench Verified | 82 | 8.7 |
| TauBench V3 - Banking | 24.5 | 2.9 |
| Terminal-Bench 2.1 | 55.8 | 5.6 |
| τ³-Bench | 19 | 2.9 |
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.