| Benchmark | Gemini 1.5 Pro | Gemma 4 26B A4B |
|---|---|---|
| AA Intelligence | 10 | 26.1 |
| AI2D | 80.9 | 88.3 |
| arena_vision | 1179 | 1238 |
| Artificial Analysis Coding Index | 23.6 | 39.3 |
| GPQA Diamond | 59.1 | 82.3 |
| GSM8K | 95.2 | 77 |
| hellaswag | 93.3 | 85.3 |
| HLE | 4.9 | 19.3 |
| humaneval | 86.6 | 50 |
| ifeval | 89.4 | 96.4 |
| LiveCodeBench | 30.5 | 79.8 |
| MathVista | 70.6 | 79.4 |
| mmlu | 86.8 | 77.8 |
| MMLU-Pro | 75.8 | 85.2 |
| MMMU | 68.4 | 78.4 |
| MMMU-Pro | 55 | 73.8 |
| OpenBookQA | 95.2 | 48.8 |
| RULER 1M | 85 | 72.9 |
| RULER 256K | 91.6 | 85.7 |
| scicode | 29.5 | 40.3 |
| SWE-bench Verified | 34.2 | 57.4 |
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.