| Benchmark | GPT-4o mini | GPT-5 |
|---|---|---|
| AA Agentic Index | 1 | 49.7 |
| AA Intelligence | 7 | 35.3 |
| aider_polyglot | 3.6 | 88 |
| AIR-Bench 2024 | 56.3 | 87.7 |
| ARC-AGI-2 | 0 | 7.5 |
| Artificial Analysis Coding Index | 11.4 | 38.9 |
| Fortress | 48.1 | 17 |
| gaia_level1 | 7.5 | 78.5 |
| gaia_level2 | 4.4 | 64.8 |
| gaia_level3 | 0 | 46.9 |
| gaia_overall | 4.7 | 66.1 |
| gdpval | 0 | 32.5 |
| GPQA Diamond | 42.6 | 87.3 |
| GSM8K | 84.3 | -1.8 |
| harmbench | 84.9 | 70 |
| HLE | 4.2 | 28.5 |
| humaneval | 87.2 | 93.4 |
| IFBench | 31 | 73.1 |
| LiveCodeBench | 35.5 | 85 |
| MMMU | 59.4 | 84.2 |
| MMMU-Pro | 41.5 | 78.4 |
| scicode | 22.9 | 43 |
| simplebench | 10.7 | 56.7 |
| SWE-bench Verified | 8.7 | 74.9 |
| TauBench V3 - Banking | 2.9 | 22.1 |
| Terminal-Bench 2.1 | 5.6 | 35.2 |
| τ³-Bench | 2.9 | 19.6 |
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.