| Benchmark | Gemini 3 Pro | o3 |
|---|---|---|
| AA Agentic Index | 52 | 36.1 |
| AA Intelligence | 40.6 | 31.1 |
| AA-LCR | 73 | 73.3 |
| AA-Omniscience | 15.3 | -15.3 |
| AIME 2025 | 100 | 89.2 |
| arc_agi_1 | 75 | 60.8 |
| ARC-AGI-2 | 54 | 6.5 |
| arena_vision | 1290 | 1217 |
| Artificial Analysis Coding Index | 46.5 | 38.4 |
| browsecomp | 59.2 | 49.7 |
| critpt | 9.1 | 1.1 |
| Fortress | 41.7 | 16 |
| frontiermath_tier_4 | 18.8 | 2.1 |
| gdpval | 34.2 | 12.8 |
| GPQA Diamond | 91.9 | 83.3 |
| HLE | 45.8 | 20.3 |
| hmmt_feb_2025 | 97.5 | 77.5 |
| IFBench | 70.4 | 71.4 |
| LiveCodeBench | 92 | 84.7 |
| longbench_v2 | 68.2 | 58.8 |
| MathVista | 89.8 | 86.8 |
| metr_hcast | 71.3 | 64.7 |
| metr_re_bench | 35 | 15 |
| metr_swaa | 99.5 | 99.8 |
| MMLU-Pro | 90.1 | 85 |
| MMMU-Pro | 81 | 76.4 |
| multichallenge | 65.7 | 60.4 |
| MultiNRC | 59 | 45.5 |
| OmniScience Accuracy | 55.8 | 38.6 |
| OmniScience Non-Hallucination | 10 | 12.9 |
| PropensityBench | 52.9 | 10.5 |
| scicode | 56.1 | 41 |
| simplebench | 76.4 | 53.1 |
| simpleqa | 72.1 | 49.4 |
| simpleqa_verified | 72.9 | 53 |
| swe_bench_bash | 74.2 | 58.4 |
| SWE-bench Verified | 78 | 69.1 |
| Terminal-Bench Hard | 56.9 | 37.1 |
| VideoMMMU | 87.6 | 83.3 |
| VisualToolBench | 26.9 | 13.7 |
| τ²-Bench Telecom (AA run) | 98 | 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.