| Benchmark | Gemma 4 31B | o3 |
|---|---|---|
| AA Agentic Index | 14.4 | 36.1 |
| AA Intelligence | 29.7 | 31.1 |
| AA-LCR | 68.3 | 73.3 |
| AA-Omniscience | -47.9 | -15.3 |
| arena_vision | 1252 | 1217 |
| Artificial Analysis Coding Index | 43.4 | 38.4 |
| critpt | 1.4 | 1.1 |
| ERQA | 57.5 | 64 |
| gdpval | 15.5 | 12.8 |
| GPQA Diamond | 85.7 | 83.3 |
| HLE | 26.5 | 20.3 |
| hmmt_feb_2025 | 88.7 | 77.5 |
| IFBench | 75.6 | 71.4 |
| MathVista | 79.3 | 86.8 |
| MMLU-Pro | 85.2 | 85 |
| MMMU | 80.4 | 82.9 |
| MMMU-Pro | 76.9 | 76.4 |
| OmniScience Accuracy | 20 | 38.6 |
| OmniScience Non-Hallucination | 18.1 | 12.9 |
| scicode | 43.4 | 41 |
| simpleqa_verified | 9.6 | 53 |
| SWE-bench Verified | 52 | 69.1 |
| Terminal-Bench Hard | 36.4 | 37.1 |
| τ²-Bench Telecom (AA run) | 65.5 | 80.7 |
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.