Early Stopping for Large Reasoning Models via Confidence Dynamics
2026-04-06 • Computation and Language
Computation and LanguageArtificial IntelligenceMachine Learning
AI summaryⓘ
The authors study how large reasoning models decide when to stop thinking and give an answer to avoid wasting time or making mistakes. They find that when the model is on the right track, it becomes confident early, but when it's wrong, confidence is low and reasoning drags on. Using this, they create CoDE-Stop, a simple way to watch confidence levels and stop reasoning at the right time without extra training. Their method saves computing resources and keeps accuracy high compared to earlier methods. They also analyze how confidence changes during reasoning to better understand the model's behavior.
large reasoning modelschain-of-thoughtearly stoppingconfidence dynamicscomputational costanswer confidencereasoning trajectoriesCoDE-Stopaccuracy-compute tradeofftoken usage
Authors
Parsa Hosseini, Sumit Nawathe, Mahdi Salmani, Meisam Razaviyayn, Soheil Feizi
Abstract
Large reasoning models rely on long chain-of-thought generation to solve complex problems, but extended reasoning often incurs substantial computational cost and can even degrade performance due to overthinking. A key challenge is determining when the model should stop reasoning and produce the final answer. In this work, we study the confidence of intermediate answers during reasoning and observe two characteristic behaviors: correct reasoning trajectories often reach high-confidence answers early, while incorrect rollouts tend to produce long, unproductive reasoning traces and exhibit less reliable confidence dynamics. Motivated by these observations, we propose CoDE-Stop (Confidence Dynamics Early Stop), an early stopping method that leverages the dynamics of intermediate answer confidence to decide when to terminate reasoning, requiring no additional training and easily integrating into existing models. We evaluate CoDE-Stop on diverse reasoning and science benchmarks across multiple models. Compared to prior early stopping methods, it achieves a more favorable accuracy-compute tradeoff and reduces total token usage by 25-50% compared to standard full-length reasoning. In addition, we provide analyses of confidence dynamics during reasoning, offering insights into how confidence changes in both correct and incorrect trajectories.