On the Optimal Reasoning Length for RL-Trained Language Models

TL;DR

This paper investigates the optimal output length for reinforcement learning-trained language models, balancing reasoning performance and computational efficiency, and identifies key failure modes related to output length.

Contribution

It compares length control methods on RL-trained models, revealing how length penalties affect reasoning and efficiency, and extends prior work to RL policies.

Findings

Length penalties may hinder reasoning acquisition.

Properly tuned length control can improve efficiency.

Long outputs increase dispersion, short outputs lead to under-thinking.

Abstract

Reinforcement learning substantially improves reasoning in large language models, but it also tends to lengthen chain of thought outputs and increase computational cost during both training and inference. Though length control methods have been proposed, it remains unclear what the optimal output length is for balancing efficiency and performance. In this work, we compare several length control methods on two models, Qwen3-1.7B Base and DeepSeek-R1-Distill-Qwen-1.5B. Our results indicate that length penalties may hinder reasoning acquisition, while properly tuned length control can improve efficiency for models with strong prior reasoning. By extending prior work to RL trained policies, we identify two failure modes, 1) long outputs increase dispersion, and 2) short outputs lead to under-thinking.