| Benchmark | Gemini 2.5 Flash | o3 |
|---|---|---|
| AA Agentic Index | 18.8 | 36.1 |
| AA Intelligence | 17.7 | 31.1 |
| AA-LCR | 65.7 | 73.3 |
| AA-Omniscience | -29.8 | -15.3 |
| aider_polyglot | 61.9 | 81.3 |
| AIME 2024 | 82.3 | 91.6 |
| AIME 2025 | 72 | 89.2 |
| AIR-Bench 2024 | 68.7 | 84.5 |
| anthropic_red_team | 98.8 | 98.3 |
| arc_agi_1 | 33.3 | 60.8 |
| ARC-AGI-2 | 2.5 | 6.5 |
| arena_vision | 1214 | 1217 |
| Artificial Analysis Coding Index | 22.2 | 38.4 |
| bbq | 97.7 | 97.9 |
| critpt | 1.4 | 1.1 |
| frontiermath_tier_4 | 4.2 | 2.1 |
| gdpval | 11.9 | 12.8 |
| GPQA Diamond | 82.8 | 83.3 |
| harmbench | 62.6 | 98.4 |
| HLE | 12.1 | 20.3 |
| hmmt_feb_2025 | 64.2 | 77.5 |
| IFBench | 50.3 | 71.4 |
| LiveCodeBench | 76.2 | 84.7 |
| livecodebench_easy | 99.1 | 99.1 |
| livecodebench_hard | 48.3 | 66 |
| livecodebench_medium | 82.5 | 89.8 |
| MMMU | 79.7 | 82.9 |
| MMMU-Pro | 69.1 | 76.4 |
| OmniScience Accuracy | 26.1 | 38.6 |
| OmniScience Non-Hallucination | 24.7 | 12.9 |
| PropensityBench | 68 | 10.5 |
| scicode | 35.9 | 41 |
| simple_safety_tests | 98 | 99 |
| simplebench | 41.2 | 53.1 |
| simpleqa | 26.9 | 49.4 |
| swe_bench_bash | 28.7 | 58.4 |
| SWE-bench Verified | 60.4 | 69.1 |
| Terminal-Bench Hard | 13.6 | 37.1 |
| VisualToolBench | 4.7 | 13.7 |
| xstest | 98.8 | 97.3 |
| τ²-Bench Telecom (AA run) | 31.6 | 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.