| Benchmark | Claude 3.5 Sonnet | Gemma 4 26B A4B |
|---|---|---|
| AA Intelligence | 10 | 26.1 |
| AI2D | 94.7 | 88.3 |
| AIME 2025 | 7.4 | 90 |
| arena_vision | 1146 | 1238 |
| Artificial Analysis Coding Index | 30.2 | 39.3 |
| C-Eval | 76.7 | 82.5 |
| GPQA Diamond | 67.2 | 82.3 |
| GSM8K | 96.9 | 77 |
| HLE | 3.9 | 19.3 |
| hmmt_feb_2025 | 1.7 | 91.7 |
| humaneval | 93.7 | 50 |
| ifeval | 90.1 | 96.4 |
| LiveCodeBench | 32.8 | 79.8 |
| LiveCodeBench v6 | 37.2 | 77.1 |
| MathVista | 67.7 | 79.4 |
| mmlu | 88.3 | 77.8 |
| mmlu_redux | 88.9 | 92.7 |
| MMLU-Pro | 78 | 85.2 |
| MMMU | 72 | 78.4 |
| MMMU-Pro | 54.7 | 73.8 |
| OpenBookQA | 97.2 | 48.8 |
| RealWorldQA | 60.1 | 72.2 |
| scicode | 36.6 | 40.3 |
| supergpqa | 48.2 | 61.4 |
| SWE-bench Verified | 50.8 | 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.