| Benchmark | Gemini 3.5 Flash | Gemma 4 31B |
|---|---|---|
| AA Agentic Index | 70.4 | 14.4 |
| AA Intelligence | 52 | 29.7 |
| AA-LCR | 81 | 68.3 |
| AA-Omniscience | 21.2 | -47.9 |
| aime_2026 | 95 | 89.2 |
| arena_elo | 1477 | 1451 |
| arena_text_factuality | 1466 | 1449 |
| Artificial Analysis Coding Index | 70.1 | 43.4 |
| charxiv_rq | 84.2 | 67.9 |
| coding_arena_elo | 1490 | 1364 |
| critpt | 13.1 | 1.4 |
| gdpval | 57.8 | 15.5 |
| GPQA Diamond | 92.2 | 85.7 |
| HLE | 42.7 | 26.5 |
| hmmt_feb_2026 | 95.5 | 77.2 |
| IFBench | 76.3 | 75.6 |
| itbenchSre | 40.3 | 37.3 |
| MCP Atlas | 83.6 | 57.2 |
| MMMU-Pro | 84.3 | 76.9 |
| mrcr_v2_8needle_128k_average | 77.3 | 66.4 |
| OmniScience Accuracy | 51.4 | 20 |
| OmniScience Non-Hallucination | 38.2 | 18.1 |
| scicode | 53.1 | 43.4 |
| simpleqa_verified | 68.4 | 9.6 |
| SWE-bench Pro | 55.1 | 35.7 |
| TauBench V3 - Banking | 32.2 | 14.8 |
| Terminal-Bench 2.0 | 76.2 | 42.9 |
| Terminal-Bench 2.1 | 78.7 | 43.4 |
| Terminal-Bench Hard | 46.2 | 36.4 |
| τ²-Bench Telecom (AA run) | 95.6 | 65.5 |
| τ³-Bench | 25.4 | 67.5 |
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.