| Benchmark | Gemini 2.5 Pro | Gemini 3.6 Flash |
|---|---|---|
| AA Agentic Index | 7.2 | 40.5 |
| AA Intelligence | 27 | 52 |
| AA-LCR | 66 | 79 |
| AA-Omniscience | -14.3 | 22.1 |
| arc_agi_1 | 41 | 34.5 |
| ARC-AGI-2 | 4.9 | 2.6 |
| arena_elo | 1446 | 1481 |
| arena_text_factuality | 1462 | 1470 |
| Artificial Analysis Coding Index | 46.7 | 69.2 |
| coding_arena_elo | 1224 | 1538 |
| critpt | 2.6 | 10.6 |
| gdpval | 8.5 | 46.1 |
| GPQA Diamond | 84.4 | 92.8 |
| HLE | 22.5 | 40.8 |
| matharena_visual_math_overall | 77.2 | 87.8 |
| MMMU-Pro | 74.9 | 83.2 |
| OmniScience Accuracy | 39 | 50 |
| OmniScience Non-Hallucination | 12.6 | 44.4 |
| scicode | 43 | 52.7 |
| simpleqa_verified | 56 | 68.7 |
| TauBench V3 - Banking | 9.7 | 29.9 |
| Terminal-Bench 2.1 | 28.5 | 78 |
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.