| Benchmark | Gemini 2.5 Flash | Gemini 3.6 Flash |
|---|---|---|
| AA Agentic Index | 18.8 | 40.5 |
| AA Intelligence | 17.7 | 52 |
| AA-LCR | 65.7 | 79 |
| AA-Omniscience | -29.8 | 22.1 |
| arc_agi_1 | 33.3 | 34.5 |
| ARC-AGI-2 | 2.5 | 2.6 |
| arena_text_factuality | 1437 | 1470 |
| Artificial Analysis Coding Index | 22.2 | 69.2 |
| critpt | 1.4 | 10.6 |
| gdpval | 11.9 | 46.1 |
| GPQA Diamond | 82.8 | 92.8 |
| HLE | 12.1 | 40.8 |
| MMMU-Pro | 69.1 | 83.2 |
| OmniScience Accuracy | 26.1 | 50 |
| OmniScience Non-Hallucination | 24.7 | 44.4 |
| scicode | 35.9 | 52.7 |
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.