Steering Large Reasoning Models towards Concise Reasoning via Flow Matching

TL;DR

FlowSteer introduces a nonlinear, distribution-based method to steer large reasoning models towards more concise outputs, outperforming linear approaches in efficiency and task performance.

Contribution

This work presents FlowSteer, a novel nonlinear distribution transformation using Flow Matching for input-dependent reasoning control in LRMs, surpassing prior linear methods.

Findings

FlowSteer achieves more compact reasoning outputs.

It demonstrates improved token efficiency over baselines.

Strong performance across diverse reasoning benchmarks.

Abstract

Large Reasoning Models (LRMs) excel at complex reasoning tasks, but their efficiency is often hampered by overly verbose outputs. Prior steering methods attempt to address this issue by applying a single, global vector to hidden representations -- an approach grounded in the restrictive linear representation hypothesis. In this work, we introduce FlowSteer, a nonlinear steering method that goes beyond uniform linear shifts by learning a complete transformation between the distributions associated with verbose and concise reasoning. This transformation is learned via Flow Matching as a velocity field, enabling precise, input-dependent control over the model's reasoning process. By aligning steered representations with the distribution of concise-reasoning activations, FlowSteer yields more compact reasoning than the linear shifts. Across diverse reasoning benchmarks, FlowSteer demonstrates strong task performance and token efficiency compared to leading inference-time baselines. Our work demonstrates that modeling the full distributional transport with generative techniques offers a more effective and principled foundation for controlling LRMs.