| Benchmark | Gemini 3 Pro | o4-mini |
|---|---|---|
| AA Agentic Index | 52 | 36.1 |
| AA Intelligence | 40.6 | 26.1 |
| AA-LCR | 73 | 60 |
| AA-Omniscience | 15.3 | -35.7 |
| AIME 2025 | 100 | 92.7 |
| arc_agi_1 | 75 | 58.7 |
| ARC-AGI-2 | 54 | 6.1 |
| arena_vision | 1290 | 1201 |
| Artificial Analysis Coding Index | 46.5 | 25.6 |
| browsecomp | 59.2 | 51.5 |
| critpt | 9.1 | 0.6 |
| Fortress | 41.7 | 21.5 |
| frontiermath_tier_4 | 18.8 | 6.3 |
| gdpval | 34.2 | 25.4 |
| GPQA Diamond | 91.9 | 81.4 |
| HLE | 45.8 | 16.5 |
| IFBench | 70.4 | 68.7 |
| LiveCodeBench | 92 | 84.5 |
| MathVista | 89.8 | 84.3 |
| MMMU-Pro | 81 | 69.2 |
| multichallenge | 65.7 | 44.9 |
| MultiNRC | 59 | 22.2 |
| OmniScience Accuracy | 55.8 | 24.8 |
| OmniScience Non-Hallucination | 10 | 19.5 |
| PropensityBench | 52.9 | 15.8 |
| scicode | 56.1 | 46.5 |
| SEAL VISTA | 51.5 | 51.8 |
| simplebench | 76.4 | 38.7 |
| simpleqa_verified | 72.9 | 23.9 |
| swe_bench_bash | 74.2 | 45 |
| SWE-bench Verified | 78 | 68.1 |
| Terminal-Bench Hard | 56.9 | 15.2 |
| vectara_answer_rate | 99.4 | 99.2 |
| vectara_avg_summary_length | 101.9 | 130.9 |
| vectara_factual_consistency | 86.4 | 81.4 |
| vectara_hallucination_rate ↓ | 13.6 | 18.6 |
| VisualToolBench | 26.9 | 11.1 |
| τ²-Bench Telecom (AA run) | 98 | 55.6 |
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.