| Benchmark | Gemini 3 Flash Preview | GPT-4o mini |
|---|---|---|
| AA Agentic Index | 49.7 | 1 |
| AA Intelligence | 38.7 | 7 |
| ARC-AGI-2 | 33.6 | 0 |
| Artificial Analysis Coding Index | 42.6 | 11.4 |
| gaia_level1 | 82.8 | 7.5 |
| gaia_level2 | 66 | 4.4 |
| gaia_level3 | 49 | 0 |
| gaia_overall | 68.4 | 4.7 |
| gdpval | 35.2 | 0 |
| GPQA Diamond | 90.4 | 42.6 |
| harmbench | 70 | 84.9 |
| HLE | 43.5 | 4.2 |
| IFBench | 78 | 31 |
| MMMU-Pro | 81.2 | 41.5 |
| scicode | 50.6 | 22.9 |
| simplebench | 61.1 | 10.7 |
| simpleqa | 68.7 | 9.9 |
| SWE-bench Verified | 78 | 8.7 |
| TauBench V3 - Banking | 20.8 | 2.9 |
| Terminal-Bench 2.1 | 58 | 5.6 |
| τ³-Bench | 17.5 | 2.9 |
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.