| Benchmark | Gemini 3.5 Flash | Gemma 4 E2B |
|---|---|---|
| AA Agentic Index | 70.4 | 7.4 |
| AA Intelligence | 52 | 10 |
| AA-LCR | 81 | 17 |
| AA-Omniscience | 21.2 | -23.6 |
| aime_2026 | 95 | 37.5 |
| Artificial Analysis Coding Index | 70.1 | 8.3 |
| critpt | 13.1 | 0 |
| gdpval | 57.8 | 0 |
| GPQA Diamond | 92.2 | 43.4 |
| HLE | 42.7 | 4.8 |
| IFBench | 76.3 | 38 |
| MMMU-Pro | 84.3 | 44.6 |
| mrcr_v2_8needle_128k_average | 77.3 | 19.1 |
| OmniScience Accuracy | 51.4 | 8.2 |
| OmniScience Non-Hallucination | 38.2 | 67.6 |
| scicode | 53.1 | 20.9 |
| TauBench V3 - Banking | 32.2 | 4.5 |
| Terminal-Bench 2.1 | 78.7 | 0.4 |
| Terminal-Bench Hard | 46.2 | 3 |
| τ²-Bench Telecom (AA run) | 95.6 | 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.