| Benchmark | Gemini 3 Flash Preview | o3 |
|---|---|---|
| AA Agentic Index | 49.7 | 36.1 |
| AA Intelligence | 38.7 | 31.1 |
| AA-LCR | 73 | 73.3 |
| AA-Omniscience | 10.1 | -15.3 |
| AIME 2025 | 99.7 | 89.2 |
| arc_agi_1 | 84.7 | 60.8 |
| ARC-AGI-2 | 33.6 | 6.5 |
| arena_vision | 1260 | 1217 |
| Artificial Analysis Coding Index | 42.6 | 38.4 |
| critpt | 8.6 | 1.1 |
| frontiermath_tier_4 | 4.2 | 2.1 |
| gdpval | 35.2 | 12.8 |
| GPQA Diamond | 90.4 | 83.3 |
| harmbench | 70 | 98.4 |
| HLE | 43.5 | 20.3 |
| hmmt_feb_2025 | 97.5 | 77.5 |
| IFBench | 78 | 71.4 |
| MMMU-Pro | 81.2 | 76.4 |
| OmniScience Accuracy | 53.4 | 38.6 |
| OmniScience Non-Hallucination | 7.6 | 12.9 |
| scicode | 50.6 | 41 |
| simplebench | 61.1 | 53.1 |
| simpleqa | 68.7 | 49.4 |
| simpleqa_verified | 67.4 | 53 |
| swe_bench_bash | 75.8 | 58.4 |
| SWE-bench Verified | 78 | 69.1 |
| Terminal-Bench Hard | 38.6 | 37.1 |
| VideoMMMU | 86.9 | 83.3 |
| τ²-Bench Telecom (AA run) | 80.4 | 80.7 |
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.