| Benchmark | Claude Opus 4.6 | Gemma 4 E2B |
|---|---|---|
| AA Agentic Index | 67.6 | 7.4 |
| AA Intelligence | 44.9 | 10 |
| AA-LCR | 74.3 | 17 |
| AA-Omniscience | 13.7 | -23.6 |
| aime_2026 | 96.7 | 37.5 |
| AIME25 no tools | 95.6 | 26.3 |
| Artificial Analysis Coding Index | 48.1 | 8.3 |
| critpt | 12.6 | 0 |
| gdpval | 55.9 | 0 |
| GPQA Diamond | 91.3 | 43.4 |
| HLE | 53.1 | 4.8 |
| IFBench | 62.5 | 38 |
| LiveCodeBench v6 | 88.8 | 44 |
| mathvision | 71.2 | 52.4 |
| MMLU-Pro | 89.1 | 60 |
| mmmlu | 91.1 | 67.4 |
| MMMU-Pro | 77.3 | 44.6 |
| mrcr_v2_8needle_128k_average | 84 | 19.1 |
| ocrbench_v2 | 48.4 | 44.7 |
| OmniScience Accuracy | 47 | 8.2 |
| OmniScience Non-Hallucination | 37.2 | 67.6 |
| RealWorldQA | 73.9 | 56.2 |
| scicode | 52 | 20.9 |
| Tau2 retail | 91.9 | 18.9 |
| Terminal-Bench Hard | 65.4 | 3 |
| τ²-Bench Telecom (AA run) | 99.3 | 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.