| Benchmark | GPT-4o mini | o3 |
|---|---|---|
| AA Agentic Index | 1 | 36.1 |
| AA Intelligence | 7 | 31.1 |
| aider_polyglot | 3.6 | 81.3 |
| AIR-Bench 2024 | 56.3 | 84.5 |
| anthropic_red_team | 98.3 | 98.3 |
| ARC-AGI-2 | 0 | 6.5 |
| Artificial Analysis Coding Index | 11.4 | 38.4 |
| bbq | 88.2 | 97.9 |
| Fortress | 48.1 | 16 |
| gdpval | 0 | 12.8 |
| GPQA Diamond | 42.6 | 83.3 |
| harmbench | 84.9 | 98.4 |
| HLE | 4.2 | 20.3 |
| IFBench | 31 | 71.4 |
| LiveCodeBench | 35.5 | 84.7 |
| livecodebench_easy | 81.7 | 99.1 |
| livecodebench_hard | 4.3 | 66 |
| livecodebench_medium | 25.2 | 89.8 |
| MathVista | 56.7 | 86.8 |
| MMMU | 59.4 | 82.9 |
| MMMU-Pro | 41.5 | 76.4 |
| scicode | 22.9 | 41 |
| simple_safety_tests | 97.8 | 99 |
| simplebench | 10.7 | 53.1 |
| simpleqa | 9.9 | 49.4 |
| SWE-bench Verified | 8.7 | 69.1 |
| xstest | 96 | 97.3 |
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.