@misc{indiciae5306e74f04fd,
  title = {SteadyThought: Mitigating LLM Under-Thinking via Thought-Level Preference Optimization},
  author = {Zhenyue Peng and Lin Yang and Jie Wang and Xiaoqi Ni and Hanzhu Chen and Xize Liang and Yanjun Wu and Rujing Wang and Kaiwen Zhou and Jianye Hao},
  year = {2026},
  url = {https://arxiv.org/abs/2610.25045},
  note = {Source identifier: 2610.25045}
}