| Benchmark | Claude 3.7 Sonnet | Gemma 4 26B A4B |
|---|---|---|
| AA Agentic Index | 37 | 32.1 |
| AA Intelligence | 27.6 | 26.1 |
| AA-LCR | 62.3 | 61.7 |
| AA-Omniscience | -0.7 | -50.8 |
| AIME 2025 | 54.8 | 90 |
| arena_vision | 1195 | 1238 |
| Artificial Analysis Coding Index | 36.4 | 39.3 |
| critpt | 0.9 | 0 |
| gdpval | 27.4 | 25.7 |
| GPQA Diamond | 84.8 | 82.3 |
| HLE | 9.7 | 19.3 |
| hmmt_feb_2025 | 31.7 | 91.7 |
| IFBench | 48.3 | 72.4 |
| ifeval | 93.2 | 96.4 |
| mmmlu | 86.1 | 86.3 |
| MMMU | 75 | 78.4 |
| MMMU-Pro | 60.1 | 73.8 |
| OmniScience Accuracy | 28.2 | 19.1 |
| OmniScience Non-Hallucination | 60 | 13.6 |
| scicode | 40.3 | 40.3 |
| SWE-bench Verified | 70.3 | 57.4 |
| Terminal-Bench Hard | 21.2 | 25 |
| τ²-Bench Telecom (AA run) | 54.7 | 43.6 |
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.