| Benchmark | Gemini 3.6 Flash | Gemini 3 Flash Preview |
|---|---|---|
| AA Agentic Index | 40.5 | 49.7 |
| AA Intelligence | 52 | 38.7 |
| AA-LCR | 79 | 73 |
| AA-Omniscience | 22.1 | 10.1 |
| aime_2026 | 96.7 | 96.7 |
| arc_agi_1 | 34.5 | 84.7 |
| ARC-AGI-2 | 2.6 | 33.6 |
| arena_elo | 1481 | 1473 |
| arena_text_factuality | 1470 | 1452 |
| Artificial Analysis Coding Index | 69.2 | 42.6 |
| charxiv_rq | 85.2 | 80.3 |
| coding_arena_elo | 1538 | 1387 |
| critpt | 10.6 | 8.6 |
| gdpval | 46.1 | 35.2 |
| GPQA Diamond | 92.8 | 90.4 |
| HLE | 40.8 | 43.5 |
| hmmt_feb_2026 | 89.4 | 89.4 |
| matharena_visual_math_overall | 87.8 | 85.8 |
| MMMU-Pro | 83.2 | 81.2 |
| mrcr_v2_8needle_128k_average | 91.8 | 67.2 |
| OmniScience Accuracy | 50 | 53.4 |
| OmniScience Non-Hallucination | 44.4 | 7.6 |
| OSWorld-Verified | 83 | 65.1 |
| scicode | 52.7 | 50.6 |
| simpleqa_verified | 68.7 | 67.4 |
| SWE-bench Pro | 58.7 | 49.6 |
| TauBench V3 - Banking | 29.9 | 20.8 |
| Terminal-Bench 2.1 | 78 | 58 |
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.