Small-LM powerd by reasoning ability with Qwen base model.