A compact but capable model that punches above its weight class for instruction-following tasks. It handles multilingual text well, with notably strong performance in Chinese and English, and holds its own on structured reasoning and code generation for its size. At 7.6B parameters, it occasionally struggles with complex multi-step reasoning where larger models would have more headroom.
| Benchmark | Score | Type | Recorded |
|---|---|---|---|
| BBH | 34.9 | accuracy | 7mo ago |
| MMLU-Pro | 36.5 | accuracy | 7mo ago |
| MuSR | 8.5 | accuracy | 7mo ago |
| MATH | 50.0 | accuracy | 7mo ago |
| GPQA Diamond | 5.5 | accuracy | 7mo ago |
| IFEval | 75.9 | accuracy | 7mo ago |