| Benchmark | Gemini 3.1 Pro | Gemma 4 E2B |
|---|---|---|
| AA Agentic Index | 23 | 7.4 |
| AA Intelligence | 47.7 | 10 |
| AA-LCR | 79 | 17 |
| AA-Omniscience | 32.9 | -23.6 |
| aime_2026 | 98.3 | 37.5 |
| Artificial Analysis Coding Index | 68.8 | 8.3 |
| Codeforces ELO | 405 | 633 |
| critpt | 17.7 | 0 |
| gdpval | 23.3 | 0 |
| GPQA Diamond | 94.3 | 43.4 |
| HLE | 51.4 | 4.8 |
| IFBench | 77.1 | 38 |
| LiveCodeBench v6 | 91.7 | 44 |
| mathvision | 89.8 | 52.4 |
| MMLU-Pro | 91 | 60 |
| mmmlu | 92.6 | 67.4 |
| MMMU-Pro | 83 | 44.6 |
| mrcr_v2_8needle_128k_average | 84.9 | 19.1 |
| OmniScience Accuracy | 55.3 | 8.2 |
| OmniScience Non-Hallucination | 50.1 | 67.6 |
| scicode | 59 | 20.9 |
| TauBench V3 - Banking | 21.4 | 4.5 |
| Terminal-Bench 2.1 | 74 | 0.4 |
| Terminal-Bench Hard | 68.5 | 3 |
| τ²-Bench Telecom (AA run) | 99.3 | 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.