| Benchmark | Gemma 4 26B A4B | GPT-4o mini |
|---|---|---|
| AA Agentic Index | 32.1 | 1 |
| AA Intelligence | 26.1 | 7 |
| AI2D | 88.3 | 75.2 |
| Artificial Analysis Coding Index | 39.3 | 11.4 |
| gdpval | 25.7 | 0 |
| GPQA Diamond | 82.3 | 42.6 |
| GSM8K | 77 | 84.3 |
| HLE | 19.3 | 4.2 |
| humaneval | 50 | 87.2 |
| IFBench | 72.4 | 31 |
| LiveCodeBench | 79.8 | 35.5 |
| MathVista | 79.4 | 56.7 |
| mmlu | 77.8 | 81.8 |
| MMMU | 78.4 | 59.4 |
| MMMU-Pro | 73.8 | 41.5 |
| OpenBookQA | 48.8 | 92 |
| scicode | 40.3 | 22.9 |
| SWE-bench Verified | 57.4 | 8.7 |
| TauBench V3 - Banking | 12 | 2.9 |
| Terminal-Bench 2.1 | 39 | 5.6 |
| τ³-Bench | 59 | 2.9 |
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.