| Benchmark | Gemma 4 31B | o1 |
|---|---|---|
| AA Agentic Index | 14.4 | 31.1 |
| AA Intelligence | 29.7 | 24 |
| AA-LCR | 68.3 | 63.3 |
| AA-Omniscience | -47.9 | -10.6 |
| arena_vision | 1252 | 1193 |
| Artificial Analysis Coding Index | 43.4 | 39.7 |
| critpt | 1.4 | 0.3 |
| gdpval | 15.5 | 11.5 |
| GPQA Diamond | 85.7 | 75.7 |
| HLE | 26.5 | 7.7 |
| IFBench | 75.6 | 70.3 |
| MathVista | 79.3 | 71.8 |
| mmmlu | 88.4 | 87.7 |
| MMMU | 80.4 | 77.6 |
| OmniScience Accuracy | 20 | 34.7 |
| OmniScience Non-Hallucination | 18.1 | 30.7 |
| scicode | 43.4 | 35.8 |
| SWE-bench Verified | 52 | 48.9 |
| Terminal-Bench Hard | 36.4 | 12.9 |
| τ²-Bench Telecom (AA run) | 65.5 | 62.6 |
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.