| Benchmark | Claude Opus 4.5 | Gemma 4 E2B |
|---|---|---|
| AA Agentic Index | 59.6 | 7.4 |
| AA Intelligence | 41.9 | 10 |
| AA-LCR | 76 | 17 |
| AA-Omniscience | 14 | -23.6 |
| aime_2026 | 95.1 | 37.5 |
| AIME25 no tools | 91 | 26.3 |
| Artificial Analysis Coding Index | 47.8 | 8.3 |
| Claw-Eval Avg | 76.6 | 53.1 |
| critpt | 4.6 | 0 |
| gdpval | 47.3 | 0 |
| GPQA Diamond | 87 | 43.4 |
| HLE | 30.8 | 4.8 |
| IFBench | 58 | 38 |
| LiveCodeBench v6 | 84.8 | 44 |
| mathvision | 77.1 | 52.4 |
| MathVista | 80.2 | 62.1 |
| MMLU-Pro | 90 | 60 |
| mmmlu | 90.8 | 67.4 |
| MMMU-Pro | 74 | 44.6 |
| MMStar | 73.2 | 57.9 |
| OCRBench | 86.5 | 0.7 |
| OmniScience Accuracy | 46.6 | 8.2 |
| OmniScience Non-Hallucination | 39 | 67.6 |
| RealWorldQA | 77 | 56.2 |
| scicode | 50 | 20.9 |
| Tau2 retail | 88.9 | 18.9 |
| Terminal-Bench Hard | 47 | 3 |
| τ²-Bench | 98.2 | 24.5 |
| τ²-Bench Telecom (AA run) | 89.5 | 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.