| Benchmark | Gemma 4 E2B | Opus 5 |
|---|---|---|
| AA Agentic Index | 7.4 | 59.2 |
| AA Intelligence | 10 | 63.1 |
| AA-LCR | 17 | 78.7 |
| AA-Omniscience | -23.6 | 37.1 |
| Artificial Analysis Coding Index | 8.3 | 78 |
| critpt | 0 | 29.1 |
| gdpval | 0 | 67.2 |
| GPQA Diamond | 43.4 | 93.7 |
| HLE | 4.8 | 64.7 |
| MMMU-Pro | 44.6 | 84.7 |
| OmniScience Accuracy | 8.2 | 60.9 |
| OmniScience Non-Hallucination | 67.6 | 40.5 |
| scicode | 20.9 | 55.7 |
| TauBench V3 - Banking | 4.5 | 44.7 |
| Terminal-Bench 2.1 | 0.4 | 89.1 |
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.