| Benchmark | Gemma 4 E2B | GPT-5.1 |
|---|---|---|
| AA Agentic Index | 7.4 | 32.2 |
| AA Intelligence | 10 | 37.5 |
| AA-LCR | 17 | 76.7 |
| AA-Omniscience | -23.6 | 5.4 |
| Artificial Analysis Coding Index | 8.3 | 49.4 |
| critpt | 0 | 4.9 |
| gdpval | 0 | 25 |
| GPQA Diamond | 43.4 | 88.1 |
| HLE | 4.8 | 28.5 |
| IFBench | 38 | 72.9 |
| LiveCodeBench v6 | 44 | 87 |
| MMLU-Pro | 60 | 87 |
| mmmlu | 67.4 | 91 |
| MMMU-Pro | 44.6 | 75.5 |
| OmniScience Accuracy | 8.2 | 37.7 |
| OmniScience Non-Hallucination | 67.6 | 48.1 |
| scicode | 20.9 | 43.3 |
| Tau2 retail | 18.9 | 77.9 |
| TauBench V3 - Banking | 4.5 | 15.9 |
| Terminal-Bench 2.1 | 0.4 | 52.4 |
| Terminal-Bench Hard | 3 | 45.5 |
| τ²-Bench | 24.5 | 82.7 |
| τ²-Bench Telecom (AA run) | 22.2 | 81.9 |
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.