| Benchmark | Gemma 4 26B A4B | o3 |
|---|---|---|
| AA Agentic Index | 32.1 | 36.1 |
| AA Intelligence | 26.1 | 31.1 |
| AA-LCR | 61.7 | 73.3 |
| AA-Omniscience | -50.8 | -15.3 |
| AIME 2025 | 90 | 89.2 |
| arena_vision | 1238 | 1217 |
| Artificial Analysis Coding Index | 39.3 | 38.4 |
| browsecomp | 26.3 | 49.7 |
| critpt | 0 | 1.1 |
| gdpval | 25.7 | 12.8 |
| GPQA Diamond | 82.3 | 83.3 |
| HLE | 19.3 | 20.3 |
| hmmt_feb_2025 | 91.7 | 77.5 |
| IFBench | 72.4 | 71.4 |
| LiveCodeBench | 79.8 | 84.7 |
| MathVista | 79.4 | 86.8 |
| MMLU-Pro | 85.2 | 85 |
| MMMU | 78.4 | 82.9 |
| MMMU-Pro | 73.8 | 76.4 |
| OmniScience Accuracy | 19.1 | 38.6 |
| OmniScience Non-Hallucination | 13.6 | 12.9 |
| scicode | 40.3 | 41 |
| SWE-bench Verified | 57.4 | 69.1 |
| Terminal-Bench Hard | 25 | 37.1 |
| τ²-Bench Telecom (AA run) | 43.6 | 80.7 |
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.