| Benchmark | GPT-4o mini | o1 |
|---|---|---|
| AA Agentic Index | 1 | 31.1 |
| AA Intelligence | 7 | 24 |
| aider_polyglot | 3.6 | 61.7 |
| AIR-Bench 2024 | 56.3 | 80 |
| anthropic_red_team | 98.3 | 98.3 |
| Artificial Analysis Coding Index | 11.4 | 39.7 |
| bbq | 88.2 | 97.3 |
| Fortress | 48.1 | 19.4 |
| gdpval | 0 | 11.5 |
| GPQA Diamond | 42.6 | 75.7 |
| GSM8K | 84.3 | 97.1 |
| harmbench | 84.9 | 96.3 |
| HLE | 4.2 | 7.7 |
| humaneval | 87.2 | 88.1 |
| IFBench | 31 | 70.3 |
| MATH | 80.2 | 96.4 |
| MathVista | 56.7 | 71.8 |
| mgsm | 87 | 90.8 |
| mmlu | 81.8 | 91.8 |
| MMMU | 59.4 | 77.6 |
| scicode | 22.9 | 35.8 |
| simple_safety_tests | 97.8 | 99 |
| simplebench | 10.7 | 41.7 |
| simpleqa | 9.9 | 47 |
| SWE-bench Verified | 8.7 | 48.9 |
| xstest | 96 | 97 |
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.