| Benchmark | Gemini 2.5 Pro | Gemini 3.5 Flash |
|---|---|---|
| AA Agentic Index | 7.2 | 70.4 |
| AA Intelligence | 27 | 52 |
| AA-LCR | 66 | 81 |
| AA-Omniscience | -14.3 | 21.2 |
| arc_agi_1 | 41 | 92.5 |
| ARC-AGI-2 | 4.9 | 72.1 |
| arena_elo | 1446 | 1477 |
| arena_text_factuality | 1462 | 1466 |
| Artificial Analysis Coding Index | 46.7 | 70.1 |
| coding_arena_elo | 1224 | 1490 |
| critpt | 2.6 | 13.1 |
| frontiermath_tier_4 | 4.2 | 14.6 |
| gdpval | 8.5 | 57.8 |
| GPQA Diamond | 84.4 | 92.2 |
| HLE | 22.5 | 42.7 |
| IFBench | 49 | 76.3 |
| matharena_visual_math_overall | 77.2 | 89.9 |
| MMMU-Pro | 74.9 | 84.3 |
| OmniScience Accuracy | 39 | 51.4 |
| OmniScience Non-Hallucination | 12.6 | 38.2 |
| scicode | 43 | 53.1 |
| simplebench | 62.4 | 76.7 |
| simpleqa_verified | 56 | 68.4 |
| TauBench V3 - Banking | 9.7 | 32.2 |
| Terminal-Bench 2.1 | 28.5 | 78.7 |
| Terminal-Bench Hard | 26.5 | 46.2 |
| τ²-Bench Telecom (AA run) | 54.1 | 95.6 |
| τ³-Bench | 9.3 | 25.4 |
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.