| Benchmark | Gemini 3 Flash Preview | Gemma 4 31B |
|---|---|---|
| AA Agentic Index | 49.7 | 14.4 |
| AA Intelligence | 38.7 | 29.7 |
| AA-LCR | 73 | 68.3 |
| AA-Omniscience | 10.1 | -47.9 |
| aime_2026 | 96.7 | 89.2 |
| arena_elo | 1473 | 1451 |
| arena_text_factuality | 1452 | 1449 |
| arena_vision | 1260 | 1252 |
| Artificial Analysis Coding Index | 42.6 | 43.4 |
| charxiv_rq | 80.3 | 67.9 |
| coding_arena_elo | 1387 | 1364 |
| critpt | 8.6 | 1.4 |
| gdpval | 35.2 | 15.5 |
| GPQA Diamond | 90.4 | 85.7 |
| HLE | 43.5 | 26.5 |
| hmmt_feb_2025 | 97.5 | 88.7 |
| hmmt_feb_2026 | 89.4 | 77.2 |
| hmmt_nov_2025 | 93.3 | 87.5 |
| IFBench | 78 | 75.6 |
| MCP Atlas | 62 | 57.2 |
| mmmlu | 91.8 | 88.4 |
| MMMU-Pro | 81.2 | 76.9 |
| mrcr_v2_8needle_128k_average | 67.2 | 66.4 |
| OmniDocBench 1.5 | 12.1 | 80.1 |
| OmniScience Accuracy | 53.4 | 20 |
| OmniScience Non-Hallucination | 7.6 | 18.1 |
| scicode | 50.6 | 43.4 |
| simpleqa_verified | 67.4 | 9.6 |
| SWE-bench Multilingual | 72.7 | 51.7 |
| SWE-bench Pro | 49.6 | 35.7 |
| SWE-bench Verified | 78 | 52 |
| TauBench V3 - Banking | 20.8 | 14.8 |
| Terminal-Bench 2.0 | 47.6 | 42.9 |
| Terminal-Bench 2.1 | 58 | 43.4 |
| Terminal-Bench Hard | 38.6 | 36.4 |
| vectara_answer_rate | 99.8 | 100 |
| vectara_avg_summary_length | 90.2 | 75.8 |
| vectara_factual_consistency | 86.5 | 92.6 |
| vectara_hallucination_rate ↓ | 13.5 | 7.4 |
| τ²-Bench Telecom (AA run) | 80.4 | 65.5 |
| τ³-Bench | 17.5 | 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.