| Benchmark | Gemini 3.6 Flash | Gemma 4 31B |
|---|---|---|
| AA Agentic Index | 40.5 | 14.4 |
| AA Intelligence | 52 | 29.7 |
| AA-LCR | 79 | 68.3 |
| AA-Omniscience | 22.1 | -47.9 |
| aime_2026 | 96.7 | 89.2 |
| arena_elo | 1481 | 1451 |
| arena_text_factuality | 1470 | 1449 |
| Artificial Analysis Coding Index | 69.2 | 43.4 |
| charxiv_rq | 85.2 | 67.9 |
| coding_arena_elo | 1538 | 1364 |
| critpt | 10.6 | 1.4 |
| gdpval | 46.1 | 15.5 |
| GPQA Diamond | 92.8 | 85.7 |
| HLE | 40.8 | 26.5 |
| hmmt_feb_2026 | 89.4 | 77.2 |
| MMMU-Pro | 83.2 | 76.9 |
| mrcr_v2_8needle_128k_average | 91.8 | 66.4 |
| OmniScience Accuracy | 50 | 20 |
| OmniScience Non-Hallucination | 44.4 | 18.1 |
| scicode | 52.7 | 43.4 |
| simpleqa_verified | 68.7 | 9.6 |
| SWE-bench Pro | 58.7 | 35.7 |
| TauBench V3 - Banking | 29.9 | 14.8 |
| Terminal-Bench 2.1 | 78 | 43.4 |
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.