A compact, conversational model that punches reasonably well for its size, with particular attention paid to instruction-following and dialogue coherence. It handles Chinese and English with notable fluency, reflecting its training origins at Shanghai AI Laboratory. Like many mid-sized open models, it can stumble on complex multi-step reasoning or highly specialized domains.
| Benchmark | Score | Type | Recorded |
|---|---|---|---|
| IFEval | 55.4 | accuracy | 6d ago |