| Benchmark | GPT-4o mini | GPT-5.2 |
|---|---|---|
| AA Agentic Index | 1 | 60.2 |
| AA Intelligence | 7 | 43.3 |
| ARC-AGI-2 | 0 | 52.9 |
| Artificial Analysis Coding Index | 11.4 | 48.7 |
| Fortress | 48.1 | 17.5 |
| gaia_level1 | 7.5 | 92.5 |
| gaia_level2 | 4.4 | 74.8 |
| gaia_level3 | 0 | 57.1 |
| gaia_overall | 4.7 | 77.4 |
| gdpval | 0 | 48.3 |
| GPQA Diamond | 42.6 | 92.4 |
| HLE | 4.2 | 45.5 |
| IFBench | 31 | 75.4 |
| LiveCodeBench | 35.5 | 89 |
| MathVista | 56.7 | 82.8 |
| MMMU-Pro | 41.5 | 79.5 |
| scicode | 22.9 | 52.1 |
| simplebench | 10.7 | 45.8 |
| SWE-bench Verified | 8.7 | 80 |
| TauBench V3 - Banking | 2.9 | 11.1 |
| Video-MME | 68.9 | 86 |
| τ³-Bench | 2.9 | 12 |
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.