| Benchmark | Claude Sonnet 4.5 | Gemma 4 E2B |
|---|---|---|
| AA Agentic Index | 50.6 | 7.4 |
| AA Intelligence | 37.4 | 10 |
| AA-LCR | 68.3 | 17 |
| AA-Omniscience | -0.1 | -23.6 |
| AIME25 no tools | 88 | 26.3 |
| Artificial Analysis Coding Index | 52.1 | 8.3 |
| critpt | 1.1 | 0 |
| gdpval | 40.4 | 0 |
| GPQA Diamond | 83.4 | 43.4 |
| HallusionBench | 59.9 | 0.6 |
| HLE | 19.8 | 4.8 |
| IFBench | 57.3 | 38 |
| LiveCodeBench v6 | 64 | 44 |
| LiveCodeBenchV6 no tools | 64 | 54.9 |
| MathVista | 79.8 | 62.1 |
| MMLU-Pro | 88.2 | 60 |
| mmmlu | 89.1 | 67.4 |
| MMMU-Pro | 68.7 | 44.6 |
| OmniScience Accuracy | 32.9 | 8.2 |
| OmniScience Non-Hallucination | 50.9 | 67.6 |
| RealWorldQA | 70.3 | 56.2 |
| scicode | 45 | 20.9 |
| TauBench V3 - Banking | 24.5 | 4.5 |
| Terminal-Bench 2.1 | 55.8 | 0.4 |
| Terminal-Bench Hard | 35.6 | 3 |
| τ²-Bench | 98 | 24.5 |
| τ²-Bench Telecom (AA run) | 78.1 | 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.