| Benchmark | Gemini 2.5 Flash | o4-mini |
|---|---|---|
| AA Agentic Index | 18.8 | 36.1 |
| AA Intelligence | 17.7 | 26.1 |
| AA-LCR | 65.7 | 60 |
| AA-Omniscience | -29.8 | -35.7 |
| aider_polyglot | 61.9 | 72 |
| AIME 2025 | 72 | 92.7 |
| AIR-Bench 2024 | 68.7 | 78.5 |
| anthropic_red_team | 98.8 | 98.2 |
| arc_agi_1 | 33.3 | 58.7 |
| ARC-AGI-2 | 2.5 | 6.1 |
| arena_vision | 1214 | 1201 |
| Artificial Analysis Coding Index | 22.2 | 25.6 |
| bbq | 97.7 | 94 |
| critpt | 1.4 | 0.6 |
| frontiermath_tier_4 | 4.2 | 6.3 |
| gdpval | 11.9 | 25.4 |
| GPQA Diamond | 82.8 | 81.4 |
| harmbench | 62.6 | 97 |
| HLE | 12.1 | 16.5 |
| IFBench | 50.3 | 68.7 |
| LiveCodeBench | 76.2 | 84.5 |
| livecodebench_easy | 99.1 | 98.8 |
| livecodebench_hard | 48.3 | 62.9 |
| livecodebench_medium | 82.5 | 92.2 |
| MMMU | 79.7 | 81.6 |
| MMMU-Pro | 69.1 | 69.2 |
| OmniScience Accuracy | 26.1 | 24.8 |
| OmniScience Non-Hallucination | 24.7 | 19.5 |
| PropensityBench | 68 | 15.8 |
| scicode | 35.9 | 46.5 |
| SEAL VISTA | 49.1 | 51.8 |
| simple_safety_tests | 98 | 100 |
| simplebench | 41.2 | 38.7 |
| swe_bench_bash | 28.7 | 45 |
| SWE-bench Verified | 60.4 | 68.1 |
| Terminal-Bench Hard | 13.6 | 15.2 |
| vectara_answer_rate | 99 | 99.2 |
| vectara_avg_summary_length | 101.5 | 130.9 |
| vectara_factual_consistency | 92.2 | 81.4 |
| vectara_hallucination_rate ↓ | 7.8 | 18.6 |
| VisualToolBench | 4.7 | 11.1 |
| xstest | 98.8 | 97.4 |
| τ²-Bench Telecom (AA run) | 31.6 | 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.