Large reasoning models that use long chain-of-thought excel at problem-solving yet waste compute on redundant checks. Curbing this overthinking is hard: training-time length penalties can cripple ability, while inference-time early-exit adds system overhead. To bridge this gap, we propose Step-GRPO, a novel post-training framework that internalizes dynamic early-exit capabilities directly into the model. Step-GRPO shifts the optimization objective from raw tokens to semantic steps by utilizing linguistic markers to structure reasoning. We introduce a Dynamic Truncated Rollout mechanism that exposes the model to concise high-confidence trajectories during exploration, synergized with a Step-Aware Relative Reward that dynamically penalizes redundancy based on group-level baselines. Extensive experiments across three model sizes on diverse benchmarks demonstrate that Step-GRPO achieves a superior accuracy-efficiency trade-off. On Qwen3-8B, our method reduces token consumption by 32.0\% compared to the vanilla model while avoiding the accuracy degradation observed in traditional length-penalty methods.
@article{arxiv.2604.16890,
title = {Step-GRPO: Internalizing Dynamic Early Exit for Efficient Reasoning},
author = {Benteng Chen and Weida Wang and Shufei Zhang and Mingbao Lin and Min Zhang},
journal= {arXiv preprint arXiv:2604.16890},
year = {2026}
}
Comments
This paper has been accepted for publication at the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026)