Step-GRPO: Internalizing Dynamic Early Exit for Efficient Reasoning
For practitioners deploying large reasoning models, Step-GRPO offers a way to reduce computational waste from overthinking while maintaining reasoning ability, addressing a key efficiency bottleneck.
Step-GRPO introduces a post-training framework that embeds dynamic early-exit into reasoning models, reducing token consumption by 32.0% on Qwen3-8B without accuracy loss, outperforming traditional length-penalty methods.
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.