| Benchmark | Claude 3.7 Sonnet | GPT-4o mini |
|---|---|---|
| AA Agentic Index | 37 | 1 |
| AA Intelligence | 27.6 | 7 |
| aider_polyglot | 64.9 | 3.6 |
| AIR-Bench 2024 | 81.8 | 56.3 |
| anthropic_red_team | 99.7 | 98.3 |
| ARC-AGI-2 | 0 | 0 |
| Artificial Analysis Coding Index | 36.4 | 11.4 |
| bbq | 92.1 | 88.2 |
| Fortress | 38 | 48.1 |
| gdpval | 27.4 | 0 |
| GPQA Diamond | 84.8 | 42.6 |
| harmbench | 84.3 | 84.9 |
| HLE | 9.7 | 4.2 |
| IFBench | 48.3 | 31 |
| MMMU | 75 | 59.4 |
| MMMU-Pro | 60.1 | 41.5 |
| scicode | 40.3 | 22.9 |
| simple_safety_tests | 100 | 97.8 |
| simplebench | 46.4 | 10.7 |
| SWE-bench Verified | 70.3 | 8.7 |
| xstest | 96.4 | 96 |
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.