| Benchmark | Gemma 4 31B | o4-mini |
|---|---|---|
| AA Agentic Index | 14.4 | 36.1 |
| AA Intelligence | 29.7 | 26.1 |
| AA-LCR | 68.3 | 60 |
| AA-Omniscience | -47.9 | -35.7 |
| arena_vision | 1252 | 1201 |
| Artificial Analysis Coding Index | 43.4 | 25.6 |
| critpt | 1.4 | 0.6 |
| gdpval | 15.5 | 25.4 |
| GPQA Diamond | 85.7 | 81.4 |
| HLE | 26.5 | 16.5 |
| IFBench | 75.6 | 68.7 |
| MathVista | 79.3 | 84.3 |
| MMMU | 80.4 | 81.6 |
| MMMU-Pro | 76.9 | 69.2 |
| OmniScience Accuracy | 20 | 24.8 |
| OmniScience Non-Hallucination | 18.1 | 19.5 |
| scicode | 43.4 | 46.5 |
| simpleqa_verified | 9.6 | 23.9 |
| SWE-bench Verified | 52 | 68.1 |
| Terminal-Bench Hard | 36.4 | 15.2 |
| vectara_answer_rate | 100 | 99.2 |
| vectara_avg_summary_length | 75.8 | 130.9 |
| vectara_factual_consistency | 92.6 | 81.4 |
| vectara_hallucination_rate ↓ | 7.4 | 18.6 |
| τ²-Bench Telecom (AA run) | 65.5 | 55.6 |
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.