| Benchmark | Gemma 4 E2B | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Agentic Index | 7.4 | 6.3 |
| AA Intelligence | 10 | 8.5 |
| AA-LCR | 17 | 24.7 |
| AA-Omniscience | -23.6 | -17.1 |
| Artificial Analysis Coding Index | 8.3 | 14.5 |
| critpt | 0 | 0 |
| gdpval | 0 | 0 |
| GPQA Diamond | 43.4 | 51.5 |
| HLE | 4.8 | 4.2 |
| IFBench | 38 | 39 |
| ifeval | 82.9 | 88.6 |
| MMLU-Pro | 60 | 73.4 |
| mmmlu | 67.4 | 73.8 |
| OmniScience Accuracy | 8.2 | 23.2 |
| OmniScience Non-Hallucination | 67.6 | 49 |
| scicode | 20.9 | 29.9 |
| Terminal-Bench Hard | 3 | 6.8 |
| τ²-Bench Telecom (AA run) | 22.2 | 19 |
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.