| Benchmark | Gemini 2.5 Pro | Gemma 4 E2B |
|---|---|---|
| AA Agentic Index | 7.2 | 7.4 |
| AA Intelligence | 27 | 10 |
| AA-LCR | 66 | 17 |
| AA-Omniscience | -14.3 | -23.6 |
| Artificial Analysis Coding Index | 46.7 | 8.3 |
| critpt | 2.6 | 0 |
| gdpval | 8.5 | 0 |
| GPQA Diamond | 84.4 | 43.4 |
| HLE | 22.5 | 4.8 |
| IFBench | 49 | 38 |
| mathvision | 73.3 | 52.4 |
| MathVista | 83.9 | 62.1 |
| MMLU-Pro | 86 | 60 |
| MMMU-Pro | 74.9 | 44.6 |
| MMStar | 79.2 | 57.9 |
| OCRBench | 85.9 | 0.7 |
| ocrbench_v2 | 59.3 | 44.7 |
| OmniScience Accuracy | 39 | 8.2 |
| OmniScience Non-Hallucination | 12.6 | 67.6 |
| scicode | 43 | 20.9 |
| TauBench V3 - Banking | 9.7 | 4.5 |
| Terminal-Bench 2.1 | 28.5 | 0.4 |
| Terminal-Bench Hard | 26.5 | 3 |
| τ²-Bench | 54 | 24.5 |
| τ²-Bench Telecom (AA run) | 54.1 | 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.