Mechanism
After training a reasoning RL policy, conducts a fusion SFT stage that mixes long-CoT trajectories (<think>...</think> triggered by /think or system flags) with fast, direct non-thinking responses (<think>\n\n</think> triggered by /no_think) on the exact same model weights, enabling seamless user-controlled or adaptive test-time compute budgeting within a single checkpoint.
Practical Notes
Eliminates the need to deploy separate Chat and Reasoner model families (e.g., V3 vs. R1, or GPT-4o vs. o3).
References
- Qwen3 Technical Report ↗
All 60 authors
An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jing Zhou, Jingren Zhou, Junyang Lin, Kai Dang, Keqin Bao, Kexin Yang, Le Yu, Lianghao Deng, Mei Li, Mingfeng Xue, Mingze Li, Pei Zhang, Peng Wang, Qin Zhu, Rui Men, Ruize Gao, Shixuan Liu, Shuang Luo, Tianhao Li, Tianyi Tang, Wenbiao Yin, Xingzhang Ren, Xinyu Wang, Xinyu Zhang, Xuancheng Ren, Yang Fan, Yang Su, Yichang Zhang, Yinger Zhang, Yu Wan, Yuqiong Liu, Zekun Wang, Zeyu Cui, Zhenru Zhang, Zhipeng Zhou, Zihan Qiu