| Benchmark | Gemma 4 26B A4B | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 32.1 | 14.4 |
| AA Intelligence | 26.1 | 7.7 |
| AA-LCR | 61.7 | 23 |
| AA-Omniscience | -50.8 | -68.4 |
| AI2D | 88.3 | 84.9 |
| AIME 2025 | 90 | 74.5 |
| Artificial Analysis Coding Index | 39.3 | 6.7 |
| critpt | 0 | 0 |
| gdpval | 25.7 | 13.8 |
| GPQA Diamond | 82.3 | 64.1 |
| HallusionBench | 66.1 | 64.1 |
| HLE | 19.3 | 4.6 |
| IFBench | 72.4 | 36.6 |
| ifeval | 96.4 | 82.6 |
| LiveCodeBench | 79.8 | 51.3 |
| LiveCodeBench v6 | 77.1 | 51.3 |
| mathvision | 82.4 | 60 |
| MathVista | 79.4 | 79.5 |
| mmlu_redux | 92.7 | 86 |
| MMLU-Pro | 85.2 | 73.6 |
| MMMU-Pro | 73.8 | 57 |
| OmniScience Accuracy | 19.1 | 12.2 |
| OmniScience Non-Hallucination | 13.6 | 8.2 |
| RealWorldQA | 72.2 | 73.2 |
| scicode | 40.3 | 17.1 |
| supergpqa | 61.4 | 46.8 |
| Terminal-Bench Hard | 25 | 1.5 |
| τ²-Bench Telecom (AA run) | 43.6 | 15.5 |
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.