Proceedings

Sprint or Delve: A Distribution-Aware Approach to Efficient Reasoning

Zehui Ling, Deshu Chen, Hongwei Zhang, Yifeng Jiao, Xin Guo, Zenglin Xu, Yuan Cheng

Download BibTeX Cite

Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence. Main Track. Pages 5864-5872.

https://doi.org/10.24963/ijcai.2026/653

Keywords

  • Natural Language Processing: Other
  • Natural Language Processing: Language models
  • Natural Language Processing: Question answering

Abstract

Reasoning chains in Large Language Models (LLMs) often exhibit heavy-tailed length distributions, yet existing efficiency methods rely on suboptimal linear penalties that suppress complex reasoning, limiting both accuracy and generalization. To address this, we first empirically observe that reasoning lengths are well approximated by a log-normal distribution, and provide an intuitive explanation for this phenomenon. Based on this insight, we propose the Powered Length Penalty (PLP), an adaptive regularizer that penalizes redundancy in short sequences while gradually reducing penalties for longer sequences, preserving deep reasoning. Trained solely on the elementary GSM8K dataset, PLP significantly improves reasoning efficiency by reducing inference costs on GSM8K and MATH500, while simultaneously enhancing accuracy on the challenging AIME2024 benchmark. Furthermore, PLP transfers effectively to diverse domains, including MMLU and GPQA. These results suggest that modeling reasoning length distributions and adapting penalties accordingly can mitigate the typical trade-off between efficiency and performance, enabling more reliable and cost-effective reasoning across tasks.

Zehui Ling, Deshu Chen, Hongwei Zhang, Yifeng Jiao, Xin Guo, Zenglin Xu, and Yuan Cheng. 2026. Sprint or Delve: A Distribution-Aware Approach to Efficient Reasoning. In Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence, IJCAI-26. IJCAI, 5864–5872. https://doi.org/10.24963/ijcai.2026/653
Welcome Future Conferences AI Journal Division IJCAI Proceedings About IJCAI IJCAI Awards