LAPO: Internalizing Reasoning Efficiency via Length-Adaptive Policy Optimization

TL;DR

LAPO introduces a reinforcement learning framework that enables reasoning models to adaptively control their reasoning depth, reducing token usage and improving accuracy on mathematical benchmarks by internalizing length-awareness.

Contribution

LAPO transforms reasoning length control into an intrinsic model capability using a two-stage reinforcement learning process, unlike prior external or post-hoc methods.

Findings

Reduces token usage by up to 40.9%.

Improves reasoning accuracy by 2.3%.

Models develop emergent resource allocation abilities.

Abstract

Large reasoning models have achieved remarkable performance through extended chain-of-thought sequences, yet this computational freedom leads to excessive token generation even for simple problems. We present Length-Adaptive Policy Optimization (LAPO), a novel framework that transforms reasoning length control from an external constraint into an intrinsic model capability. Unlike existing approaches that impose rigid limits or rely on post-hoc interventions, LAPO enables models to internalize an understanding of appropriate reasoning depth through a two-stage reinforcement learning process. In the first stage, models learn natural reasoning patterns by discovering the statistical distribution of successful solution lengths. The second stage leverages these patterns as meta-cognitive guidance, embedding them directly within the model's reasoning context to ensure inference-time flexibility. Experiments on mathematical reasoning benchmarks demonstrate that LAPO reduces token usage by up to 40.9% while improving accuracy by 2.3%. Our analysis reveals that models trained with LAPO develop emergent abilities to allocate computational resources based on problem complexity, achieving efficient reasoning without sacrificing quality.