Think Just Enough: Sequence-Level Entropy as a Confidence Signal for LLM Reasoning

ICLR 2026 Conference Submission17534 Authors

Keywords: adaptive compute, entropy-based early stopping, token efficiency, confidence calibration, chain-of-thought, few-shot thresholding, budget reallocation, reasoning LLMs

TL;DR: Use Shannon-entropy from token logprobs to stop CoT early, cutting 25–50% tokens/latency with no accuracy loss. Few examples set thresholds; harder items get extra budget, transferring across reasoning LLMs.

Abstract: We introduce a simple, yet novel entropy-based framework to drive token efficiency in large language models during reasoning tasks. Our approach uses Shannon entropy from token-level logprobs as a confidence signal to enable early stopping, achieving 25-50% computational savings while maintaining task accuracy. Crucially, we demonstrate that entropy-based confidence calibration represents an emergent property of advanced post-training optimization present in modern reasoning models but notably absent in standard instruction-tuned and pre-trained models (Llama 3.3 70B). We show that the entropy threshold to stop reasoning varies from model to model but can be calculated easily in one shot using only a few examples from existing reasoning datasets. Our results indicate that advanced reasoning models often know that they’ve gotten a correct answer early on, and that this emergent confidence awareness can be exploited to save tokens and reduce latency. The framework demonstrates consistent performance across reasoning-optimized model families with 25-50% computational cost reduction while preserving accuracy, revealing that confidence mechanisms represent a distinguishing characteristic of modern post-trained reasoning systems versus their predecessors.

Primary Area: foundation or frontier models, including LLMs

Submission Number: 17534