| Benchmark | DeepSeek-V4-Pro | Gemma 4 E2B |
|---|---|---|
| AA Agentic Index | 63.3 | 7.4 |
| AA Intelligence | 53 | 10 |
| AA-LCR | 70 | 17 |
| AA-Omniscience | -10.6 | -23.6 |
| aime_2026 | 96.7 | 37.5 |
| Artificial Analysis Coding Index | 59.4 | 8.3 |
| critpt | 13 | 0 |
| gdpval | 49 | 0 |
| GPQA Diamond | 90.5 | 43.4 |
| HLE | 48.2 | 4.8 |
| IFBench | 76.5 | 38 |
| MMLU-Pro | 87.5 | 60 |
| OmniScience Accuracy | 43 | 8.2 |
| OmniScience Non-Hallucination | 12.2 | 67.6 |
| scicode | 50 | 20.9 |
| TauBench V3 - Banking | 30.1 | 4.5 |
| Terminal-Bench 2.1 | 72.1 | 0.4 |
| Terminal-Bench Hard | 46.2 | 3 |
| τ²-Bench Telecom (AA run) | 96.2 | 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.