| Benchmark | Gemini 3 Pro | Gemma 4 E2B |
|---|---|---|
| AA Agentic Index | 52 | 7.4 |
| AA Intelligence | 40.6 | 10 |
| AA-LCR | 73 | 17 |
| AA-Omniscience | 15.3 | -23.6 |
| aime_2026 | 91.7 | 37.5 |
| AIME25 no tools | 96 | 26.3 |
| Artificial Analysis Coding Index | 46.5 | 8.3 |
| critpt | 9.1 | 0 |
| gdpval | 34.2 | 0 |
| GPQA Diamond | 91.9 | 43.4 |
| HLE | 45.8 | 4.8 |
| IFBench | 70.4 | 38 |
| LiveCodeBench v6 | 90.7 | 44 |
| mathvision | 86.1 | 52.4 |
| MathVista | 89.8 | 62.1 |
| MMLU-Pro | 90.1 | 60 |
| mmmlu | 91.8 | 67.4 |
| MMMU-Pro | 81 | 44.6 |
| mrcr_v2_8needle_128k_average | 77 | 19.1 |
| OCRBench | 90.3 | 0.7 |
| ocrbench_v2 | 63.4 | 44.7 |
| OmniScience Accuracy | 55.8 | 8.2 |
| OmniScience Non-Hallucination | 10 | 67.6 |
| scicode | 56.1 | 20.9 |
| Terminal-Bench Hard | 56.9 | 3 |
| τ²-Bench | 98 | 24.5 |
| τ²-Bench Telecom (AA run) | 98 | 22.2 |
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.