When Is Thinking Enough? Early Exit via Sufficiency Assessment for Efficient Reasoning

2026-04-08

AI Summary

This work addresses the issue of computational redundancy in large reasoning models caused by overthinking, a problem exacerbated by existing early-stopping methods that rely on unreliable handcrafted heuristics. To overcome this limitation, the authors propose the Dynamic Thought Sufficiency Reasoning (DTSR) framework, which for the first time integrates human-inspired metacognitive mechanisms into early-stopping decisions. DTSR employs a two-stage process—monitoring reflective signals and evaluating thought sufficiency—to dynamically assess whether the current reasoning chain is adequate for reaching a correct answer and adaptively terminates inference when sufficient. Implemented as an end-to-end system based on Qwen3, DTSR reduces reasoning length by 28.9%–34.9% with negligible performance degradation, significantly enhancing computational efficiency while effectively mitigating overthinking.

Abstract

Large reasoning models (LRMs) have achieved remarkable performance in complex reasoning tasks, driven by their powerful inference-time scaling capability. However, LRMs often suffer from overthinking, which results in substantial computational redundancy and significantly reduces efficiency. Early-exit methods aim to mitigate this issue by terminating reasoning once sufficient evidence has been generated, yet existing approaches mostly rely on handcrafted or empirical indicators that are unreliable and impractical. In this work, we introduce Dynamic Thought Sufficiency in Reasoning (DTSR), a novel framework for efficient reasoning that enables the model to dynamically assess the sufficiency of its chain-of-thought (CoT) and determine the optimal point for early exit. Inspired by human metacognition, DTSR operates in two stages: (1) Reflection Signal Monitoring, which identifies reflection signals as potential cues for early exit, and (2) Thought Sufficiency Check, which evaluates whether the current CoT is sufficient to derive the final answer. Experimental results on the Qwen3 models show that DTSR reduces reasoning length by 28.9%-34.9% with minimal performance loss, effectively mitigating overthinking. We further discuss overconfidence in LRMs and self-evaluation paradigms, providing valuable insights for early-exit reasoning.

Problem

• overthinking
• early exit
• reasoning efficiency
• large reasoning models
• computational redundancy

Innovation

• early exit
• chain-of-thought reasoning
• thought sufficiency
• dynamic assessment
• large reasoning models