English

On-Policy Supervised Fine-Tuning for Efficient Reasoning

Artificial Intelligence 2026-02-17 v1

Abstract

Large reasoning models (LRMs) are commonly trained with reinforcement learning (RL) to explore long chain-of-thought reasoning, achieving strong performance at high computational cost. Recent methods add multi-reward objectives to jointly optimize correctness and brevity, but these complex extensions often destabilize training and yield suboptimal trade-offs. We revisit this objective and challenge the necessity of such complexity. Through principled analysis, we identify fundamental misalignments in this paradigm: KL regularization loses its intended role when correctness and length are directly verifiable, and group-wise normalization becomes ambiguous under multiple reward signals. By removing these two items and simplifying the reward to a truncation-based length penalty, we show that the optimization problem reduces to supervised fine-tuning on self-generated data filtered for both correctness and conciseness. We term this simplified training strategy on-policy SFT. Despite its simplicity, on-policy SFT consistently defines the accuracy-efficiency Pareto frontier. It reduces CoT length by up to 80 while maintaining original accuracy, surpassing more complex RL-based methods across five benchmarks. Furthermore, it significantly enhances training efficiency, reducing GPU memory usage by 50% and accelerating convergence by 70%. Our code is available at https://github.com/EIT-NLP/On-Policy-SFT.

Keywords

Cite

@article{arxiv.2602.13407,
  title  = {On-Policy Supervised Fine-Tuning for Efficient Reasoning},
  author = {Anhao Zhao and Ziyang Chen and Junlong Tong and Yingqi Fan and Fanghua Ye and Shuhao Li and Yunpu Ma and Wenjie Li and Xiaoyu Shen},
  journal= {arXiv preprint arXiv:2602.13407},
  year   = {2026}
}
R2 v1 2026-07-01T10:36:10.540Z