| Benchmark | Gemini 2.5 Pro | Gemini 3 Flash Preview |
|---|---|---|
| AA Agentic Index | 7.2 | 49.7 |
| AA Intelligence | 27 | 38.7 |
| AA-LCR | 66 | 73 |
| AA-Omniscience | -14.3 | 10.1 |
| AIME 2025 | 88.3 | 99.7 |
| arc_agi_1 | 41 | 84.7 |
| ARC-AGI-2 | 4.9 | 33.6 |
| arena_elo | 1446 | 1473 |
| arena_text_factuality | 1462 | 1452 |
| arena_vision | 1246 | 1260 |
| Artificial Analysis Coding Index | 46.7 | 42.6 |
| coding_arena_elo | 1224 | 1387 |
| critpt | 2.6 | 8.6 |
| frontiermath_tier_4 | 4.2 | 4.2 |
| gdpval | 8.5 | 35.2 |
| GPQA Diamond | 84.4 | 90.4 |
| harmbench | 65.4 | 70 |
| HLE | 22.5 | 43.5 |
| hmmt_feb_2025 | 82.5 | 97.5 |
| hmmt_nov_2025 | 80 | 93.3 |
| IFBench | 49 | 78 |
| matharena_visual_math_overall | 77.2 | 85.8 |
| MMMU-Pro | 74.9 | 81.2 |
| OmniScience Accuracy | 39 | 53.4 |
| OmniScience Non-Hallucination | 12.6 | 7.6 |
| scicode | 43 | 50.6 |
| simplebench | 62.4 | 61.1 |
| simpleqa | 50.8 | 68.7 |
| simpleqa_verified | 56 | 67.4 |
| swe_bench_bash | 53.6 | 75.8 |
| SWE-bench Verified | 67.2 | 78 |
| TauBench V3 - Banking | 9.7 | 20.8 |
| Terminal-Bench 2.1 | 28.5 | 58 |
| Terminal-Bench Hard | 26.5 | 38.6 |
| vectara_answer_rate | 99.1 | 99.8 |
| vectara_avg_summary_length | 106.4 | 90.2 |
| vectara_factual_consistency | 93 | 86.5 |
| vectara_hallucination_rate ↓ | 7 | 13.5 |
| τ²-Bench Telecom (AA run) | 54.1 | 80.4 |
| τ³-Bench | 9.3 | 17.5 |
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.