| Benchmark | DeepSeek-R1 | Gemma 4 E2B |
|---|---|---|
| AA Agentic Index | 3.1 | 7.4 |
| AA Intelligence | 18.6 | 10 |
| AA-LCR | 56 | 17 |
| AA-Omniscience | -31.3 | -23.6 |
| Artificial Analysis Coding Index | 24.6 | 8.3 |
| critpt | 0.6 | 0 |
| gdpval | 1.5 | 0 |
| GPQA Diamond | 81 | 43.4 |
| HLE | 17.7 | 4.8 |
| IFBench | 39 | 38 |
| ifeval | 83.3 | 82.9 |
| MMLU-Pro | 85 | 60 |
| OmniScience Accuracy | 30.7 | 8.2 |
| OmniScience Non-Hallucination | 10.5 | 67.6 |
| scicode | 35.7 | 20.9 |
| TauBench V3 - Banking | 6.4 | 4.5 |
| Terminal-Bench 2.1 | 19.1 | 0.4 |
| Terminal-Bench Hard | 6.1 | 3 |
| τ²-Bench Telecom (AA run) | 11.4 | 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.