| Benchmark | Gemini 3.6 Flash | Gemini 3 Pro |
|---|---|---|
| AA Agentic Index | 40.5 | 52 |
| AA Intelligence | 52 | 40.6 |
| AA-LCR | 79 | 73 |
| AA-Omniscience | 22.1 | 15.3 |
| aime_2026 | 96.7 | 91.7 |
| arc_agi_1 | 34.5 | 75 |
| ARC-AGI-2 | 2.6 | 54 |
| arena_elo | 1481 | 1485 |
| arena_text_factuality | 1470 | 1481 |
| Artificial Analysis Coding Index | 69.2 | 46.5 |
| charxiv_rq | 85.2 | 81.4 |
| coding_arena_elo | 1538 | 1438 |
| critpt | 10.6 | 9.1 |
| gdpval | 46.1 | 34.2 |
| GPQA Diamond | 92.8 | 91.9 |
| HLE | 40.8 | 45.8 |
| hmmt_feb_2026 | 89.4 | 86.4 |
| LVBench | 84.2 | 73.5 |
| matharena_visual_math_overall | 87.8 | 84.2 |
| MMMU-Pro | 83.2 | 81 |
| mrcr_v2_8needle_128k_average | 91.8 | 77 |
| OmniScience Accuracy | 50 | 55.8 |
| OmniScience Non-Hallucination | 44.4 | 10 |
| scicode | 52.7 | 56.1 |
| simpleqa_verified | 68.7 | 72.9 |
| SWE-bench Pro | 58.7 | 43.3 |
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.