@inproceedings{wu-etal-2026-anti,
    title = "Anti-Length Shift: Dynamic Outlier Truncation for Training Efficient Reasoning Models",
    author = "Wu, Wei  and
      Chen, Liyi  and
      Xiao, Congxi  and
      Wang, Tianfu  and
      Wang, Qimeng  and
      Lu, Chengqiang  and
      Gao, Yan  and
      Yiwu  and
      Hu, Yao  and
      Xiong, Hui",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Proceedings of the 64th Annual Meeting of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl/2026.acl-long.1047/",
    pages = "22863--22891",
    ISBN = "979-8-89176-390-6",
    abstract = "Large reasoning models enhanced by reinforcement learning with verifiable rewards have achieved significant performance gains by extending their chain-of-thought. However, this paradigm incurs substantial deployment costs as models often exhibit excessive verbosity on simple queries. Existing efficient reasoning methods relying on explicit length penalties often introduce optimization conflicts and leave the generative mechanisms driving overthinking largely unexamined. In this paper, we identify a phenomenon termed length shift where models increasingly generate unnecessary reasoning on trivial inputs during training. To address this, we introduce Dynamic Outlier Truncation (DOT), a training-time intervention that selectively suppresses redundant tokens. This method targets only the extreme tail of response lengths within fully correct rollout groups while preserving long-horizon reasoning capabilities for complex problems. To complement this intervention and ensure stable convergence, we further incorporate auxiliary KL regularization and predictive dynamic sampling. Experimental results across multiple model scales demonstrate that our approach significantly pushes the efficiency-performance Pareto frontier outward. Notably, on the AIME-24, our method reduces inference token usage by 78{\%} while simultaneously increasing accuracy compared to the initial policy and surpassing state-of-the-art efficient reasoning methods."
}