Junyu Zhang ^∗  Yifan Sun ^∗  Tianang Leng ^∗  Jingyan Shen ^∗ 
Liu Ziyin ^†  Paul Pu Liang ^†  Huan Zhang ^† 
University of Illinois Urbana-Champaign  Massachusetts Institute of Technology
 University of Pennsylvania  New York University  NTT Research
^∗ Equal contribution ^† Equal mentorship

• [2025/11] LoRe was selected as a Best Paper Nomination at the NeurIPS 2025 Workshop on Efficient Reasoning.

Despite the superior performance of Large Reasoning Models (LRMs), their reasoning behaviors are often counterintuitive, leading to suboptimal reasoning capabilities.

We present the Laws of Reasoning (LoRe), a unified framework that characterizes intrinsic reasoning patterns in LRMs. LoRe introduces the compute law with the supplementary accuracy law, examined through two properties: monotonicity and compositionality. LoRe-Bench, our proposed benchmark, systematically measures these two tractable properties for LRMs. To address the compositionality gap observed in existing models, we develop an effective finetuning approach that enforces compute-law compositionality.

As a comprehensive study from theoretical hypotheses to empirical validation, we advance a theoretical perspective grounded in human reasoning for improving reasoning in LRMs. We hope LoRe can inspire more potential strategies that guide models toward their optimal paradigms of thinking.

🚧 Code release under construction — stay tuned! 🚧

Our SFT-Compo models are available on Hugging Face 🤗.

If you have any questions related to the code or the paper, feel free to email Junyu Zhang ([email protected]).

If you find our work useful in your research, please consider citing LoRe:

@article{LoRe25,
  title={When Reasoning Meets Its Laws},
  author={Zhang, Junyu and Sun, Yifan and Leng, Tianang and Shen, Jingyan and Ziyin, Liu and Liang, Paul Pu and Zhang, Huan},
  journal={arXiv preprint arXiv:2512.17901},
  year={2025}
}