| Benchmark | Gemini 2.5 Pro | o3 |
|---|---|---|
| AA Agentic Index | 7.2 | 36.1 |
| AA Intelligence | 27 | 31.1 |
| AA-LCR | 66 | 73.3 |
| AA-Omniscience | -14.3 | -15.3 |
| aider_polyglot | 82.2 | 81.3 |
| AIME 2025 | 88.3 | 89.2 |
| AIR-Bench 2024 | 73.6 | 84.5 |
| anthropic_red_team | 99.5 | 98.3 |
| arc_agi_1 | 41 | 60.8 |
| ARC-AGI-2 | 4.9 | 6.5 |
| arena_vision | 1246 | 1217 |
| Artificial Analysis Coding Index | 46.7 | 38.4 |
| bbq | 96.4 | 97.9 |
| browsecomp | 9.9 | 49.7 |
| critpt | 2.6 | 1.1 |
| Fortress | 54.9 | 16 |
| frontiermath_tier_4 | 4.2 | 2.1 |
| gdpval | 8.5 | 12.8 |
| GPQA Diamond | 84.4 | 83.3 |
| harmbench | 65.4 | 98.4 |
| HLE | 22.5 | 20.3 |
| hmmt_feb_2025 | 82.5 | 77.5 |
| IFBench | 49 | 71.4 |
| imo_2025 | 31.6 | 16.7 |
| LiveCodeBench | 82.7 | 84.7 |
| livecodebench_easy | 98.8 | 99.1 |
| livecodebench_hard | 59.4 | 66 |
| livecodebench_medium | 90.6 | 89.8 |
| MathVista | 83.9 | 86.8 |
| MMLU-Pro | 86 | 85 |
| MMMU | 83.9 | 82.9 |
| MMMU-Pro | 74.9 | 76.4 |
| multichallenge | 53.6 | 60.4 |
| OmniScience Accuracy | 39 | 38.6 |
| OmniScience Non-Hallucination | 12.6 | 12.9 |
| PropensityBench | 79 | 10.5 |
| scicode | 43 | 41 |
| simple_safety_tests | 97 | 99 |
| simplebench | 62.4 | 53.1 |
| simpleqa | 50.8 | 49.4 |
| simpleqa_verified | 56 | 53 |
| swe_bench_bash | 53.6 | 58.4 |
| SWE-bench Verified | 67.2 | 69.1 |
| Terminal-Bench Hard | 26.5 | 37.1 |
| xstest | 98.7 | 97.3 |
| τ²-Bench Telecom (AA run) | 54.1 | 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.