| Benchmark | Gemma 4 26B A4B | o1 |
|---|---|---|
| AA Agentic Index | 32.1 | 31.1 |
| AA Intelligence | 26.1 | 24 |
| AA-LCR | 61.7 | 63.3 |
| AA-Omniscience | -50.8 | -10.6 |
| AIME 2025 | 90 | 81.7 |
| arena_vision | 1238 | 1193 |
| Artificial Analysis Coding Index | 39.3 | 39.7 |
| critpt | 0 | 0.3 |
| gdpval | 25.7 | 11.5 |
| GPQA Diamond | 82.3 | 75.7 |
| GSM8K | 77 | 97.1 |
| HLE | 19.3 | 7.7 |
| humaneval | 50 | 88.1 |
| IFBench | 72.4 | 70.3 |
| MathVista | 79.4 | 71.8 |
| mmlu | 77.8 | 91.8 |
| mmmlu | 86.3 | 87.7 |
| MMMU | 78.4 | 77.6 |
| OmniScience Accuracy | 19.1 | 34.7 |
| OmniScience Non-Hallucination | 13.6 | 30.7 |
| scicode | 40.3 | 35.8 |
| SWE-bench Verified | 57.4 | 48.9 |
| Terminal-Bench Hard | 25 | 12.9 |
| τ²-Bench Telecom (AA run) | 43.6 | 62.6 |
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.