@inproceedings{yang-etal-2026-loopcoder,
    title = "{L}oop{C}oder: Scaling Code Intelligence via Looped Language Models",
    author = "Yang, Jian  and
      Zhang, Wei  and
      Guo, Shuyue  and
      LI, Yizhi  and
      Chai, Linzheng  and
      Ye, Zhengmao  and
      Liu, Shukai  and
      Song, Yuyang  and
      Wu, Jiajun  and
      Liu, Che  and
      Zheng, Tianyu  and
      Wu, Siwei  and
      L, Leo  and
      Ma, Xudong  and
      Hao, Chuan  and
      Tao, Ran  and
      Xing, Yan  and
      Wang, Jianzhou  and
      Tang, Mingjie  and
      Liu, Aishan  and
      Li, Zhoujun  and
      Liu, Xianglong  and
      Lv, Weifeng  and
      Dai, Bryan",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl/2026.findings-acl.796/",
    pages = "16209--16223",
    ISBN = "979-8-89176-395-1",
    abstract = "While large language models (LLMs) have mastered syntax-level code generation, complex algorithmic reasoning remains a challenge, typically addressed by scaling model depth and parameter count. Universal Transformers (UT) offer a compelling alternative by introducing a recurrent inductive bias that aligns with the recursive nature of programming logic. However, training looped architectures at scale has historically been hindered by severe instability and optimization difficulties associated with backpropagation through time (BPTT). We present LoopCoder (40B-A80B) pre-trained on 12T+ code and general tokens, along with LoopCoder-Thinking and LoopCoder-Instruct variants{---}the first large-scale looped transformer for code, achieving comparable performance to standard dense architectures with more parameters. Unlike prior approaches that restrict recurrence to small-scale tasks, we implement a comprehensive looped training protocol spanning both pre-training and post-training phases. We initiate the model via dense-to-loop transformation, folding a pre-trained dense checkpoint to initialize a recurrent block, followed by rigorous looped pre-training and specialized post-training for instruction following and reasoning. Our results establish a robust recipe for scaling coding intelligence via recurrent computation, proving that dense checkpoints serve as an optimal foundation for evolving into dynamic, looped reasoners."
}