| Benchmark | Gemini 3 Flash Preview | o4-mini |
|---|---|---|
| AA Agentic Index | 49.7 | 36.1 |
| AA Intelligence | 38.7 | 26.1 |
| AA-LCR | 73 | 60 |
| AA-Omniscience | 10.1 | -35.7 |
| AIME 2025 | 99.7 | 92.7 |
| arc_agi_1 | 84.7 | 58.7 |
| ARC-AGI-2 | 33.6 | 6.1 |
| arena_vision | 1260 | 1201 |
| Artificial Analysis Coding Index | 42.6 | 25.6 |
| critpt | 8.6 | 0.6 |
| frontiermath_tier_4 | 4.2 | 6.3 |
| gdpval | 35.2 | 25.4 |
| GPQA Diamond | 90.4 | 81.4 |
| harmbench | 70 | 97 |
| HLE | 43.5 | 16.5 |
| IFBench | 78 | 68.7 |
| MMMU-Pro | 81.2 | 69.2 |
| OmniScience Accuracy | 53.4 | 24.8 |
| OmniScience Non-Hallucination | 7.6 | 19.5 |
| scicode | 50.6 | 46.5 |
| simplebench | 61.1 | 38.7 |
| simpleqa_verified | 67.4 | 23.9 |
| swe_bench_bash | 75.8 | 45 |
| SWE-bench Verified | 78 | 68.1 |
| Terminal-Bench Hard | 38.6 | 15.2 |
| vectara_answer_rate | 99.8 | 99.2 |
| vectara_avg_summary_length | 90.2 | 130.9 |
| vectara_factual_consistency | 86.5 | 81.4 |
| vectara_hallucination_rate ↓ | 13.5 | 18.6 |
| τ²-Bench Telecom (AA run) | 80.4 | 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.