AI summaryⓘ
The authors study how large reasoning models (LRMs) express their confidence in their answers, a task known as faithful calibration (FC). They find that current models struggle to accurately match their expressed confidence with their actual certainty, especially when producing long, complex reasoning steps. To better measure this, the authors create a new method that looks at different internal signals from the models and controls for variations in reasoning paths. Their results show that expressing confidence faithfully is hard for these models and that popular ways to improve confidence expression don't always work as expected. This highlights that trustworthy confidence communication is an important, yet unresolved, challenge for deploying LRMs safely.
faithful calibrationlarge reasoning modelsconfidence expressionchain-of-thought reasoningtoken probabilitieshidden statesresponse consistencyprefix-conditioned samplingmodel uncertaintymodel alignment
Authors
Areeb Gani, Asal Meskin, Gabrielle Kaili-May Liu, Arman Cohan
Abstract
Reliable uncertainty communication is critical to the trustworthiness of LLMs, yet faithful calibration (FC)--the alignment between models' intrinsic and (linguistically) expressed confidence--is a persistent failure mode. This challenge is key for large reasoning models (LRMs), whose extended reasoning traces are often interpreted by users as evidence of deliberation, competence, and confidence. Despite the importance of FC and wide usage of LRMs, the extent to which LRMs can faithfully express their confidence remains poorly understood. Moreover, the prevailing paradigm to measure FC does not generalize well to the long chain-of-thought outputs generated by LRMs, which tend to lack clear step boundaries, involve inconsistent step structure, and encode complex conditional dependencies throughout the trace--complicating estimation of intrinsic confidence. To address this challenge, we introduce a novel framework to systematically quantify FC of LRMs. Our framework analyzes linguistic decisiveness relative to three sources of internal uncertainty, based on token probabilities, hidden states, and sampled response consistency. We also devise a prefix-conditioned sampling approach to control for conditional and structural variation across traces. Applying our framework to a diverse suite of leading models, datasets, and prompts, we find that faithful confidence expression is a significant challenge for LRMs. Reasoning behaviors do not automatically translate to improved FC, and prompt interventions for non-reasoning models do not improve faithfulness in the reasoning setting. Different confidence estimators further produce divergent assessments of the same traces, revealing fragility in prior evaluation methodologies. Taken together, our work establishes FC as a distinct reliability and alignment target for LRMs, particularly as such systems are increasingly deployed in high-stakes contexts.