DyCon: Dynamic Reasoning Control via Evolving Difficulty Modeling

2026-06-05

AI Summary

This work addresses the inefficiency of large reasoning models in handling complex tasks, which often stems from redundant reasoning—commonly referred to as “overthinking”—and the inability of existing approaches to dynamically adapt to evolving task difficulty during inference. The authors propose a training-free, general-purpose framework that, for the first time, reveals how task difficulty dynamically evolves throughout the reasoning process and is linearly encoded in step-level latent embeddings. Leveraging this insight, the method dynamically models task difficulty by analyzing these embeddings in real time, enabling adaptive control over reasoning depth. Evaluated across four models ranging from 4B to 32B parameters and twelve benchmarks spanning mathematical reasoning, question answering, and code generation, the approach significantly reduces redundant steps while maintaining or even improving accuracy and generalization performance.

Abstract

Recent advances in Large Reasoning Models (LRMs) demonstrate remarkable performance improvements by iteratively reflecting, exploring, and executing complex tasks, yet suffer from inefficiencies due to redundant reasoning, known as "overthinking". Existing methods to mitigate this issue either rely on static difficulty estimates or require task-specific training, and thus fail to adapt to the dynamic complexity during reasoning. In this work, we empirically show that the problem difficulty evolves dynamically throughout the reasoning process and is linearly encoded in the LRM's step-level embeddings. Building on this insight, we propose DyCon, a training-free framework that leverages latent step-level representations to explicitly model the evolving task difficulty, enabling the dynamic control of reasoning depth to mitigate the overthinking issue. Extensive experiments conducted on four models ranging from 4B to 32B, and across twelve benchmarks in math reasoning, general question answering, and coding tasks demonstrate that DyCon significantly enhances reasoning efficiency by reducing redundant steps without sacrificing accuracy or generalization. Project page and code are available at https://github.com/yu-lin-li/DyCon.

Problem

• overthinking
• reasoning efficiency
• dynamic difficulty
• Large Reasoning Models
• redundant reasoning

Innovation

• Dynamic Reasoning Control
• Evolving Difficulty Modeling
• Overthinking Mitigation
• Training-Free Framework
• Step-Level Embeddings