Quantifying Faithful Confidence Expression in Large Reasoning Models
On the map: LLM reasoning → Test-time compute and efficient thinking
Abstract (arXiv)
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.
Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 32 of its 71 references that are in the map. First posted on arXiv 2 Jun 2026.
Builds on
32 of 71 references.
Closest ancestors · 8
Same direction, newest first
- 2026The Path of Least Resistance: Guiding LLM Reasoning Trajectories with Prefix Consensus29 Jan
- 2026Confidence over Time: Confidence Calibration with Temporal Logic for Large Language Model Reasoning19 Jan
- 2026Recurrent Confidence Chain: Temporal-Aware Uncertainty Quantification in Large Language Models19 Jan
- 2025Deep Think with Confidence21 Aug
- 2025Verbalized Confidence Triggers Self-Verification: Emergent Behavior Without Explicit Reasoning Supervision4 Jun
- 2025Reasoning Models Better Express Their Confidence20 May
- 2025CER: Confidence Enhanced Reasoning in LLMs20 Feb
- 2022Language Models (Mostly) Know What They Know11 Jul
Other references · 24
- 2024The Llama 3 Herd of Modelscited by 14,715
- 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
- 2022Teaching Models to Express Their Uncertainty in Wordscited by 751
- 2023Measuring Faithfulness in Chain-of-Thought Reasoningcited by 703
- 2025Humanity's Last Examcited by 491
- 2025Reasoning Models Don't Always Say What They Thinkcited by 415
- 2024A Comprehensive Survey of Hallucination Mitigation Techniques in Large Language Modelscited by 269
- 2023MuSR: Testing the Limits of Chain-of-thought with Multistep Soft Reasoningcited by 201
- 2025SuperGPQA: Scaling LLM Evaluation across 285 Graduate Disciplinescited by 79
- 2025Calibrating Verbal Uncertainty as a Linear Feature to Reduce Hallucinationscited by 48
- 2025Evaluating Large Language Models in Scientific Discoverycited by 42
- 2025ConfTuner: Training Large Language Models to Express Their Confidence Verballycited by 39
12 more
- 2025Trust Me, I'm Wrong: LLMs Hallucinate with Certainty Despite Knowing the Answercited by 27
- 2025MetaFaith: Faithful Natural Language Uncertainty Expression in LLMscited by 16
- 2025Thought Branches: Interpreting LLM Reasoning Requires Resamplingcited by 16
- 2025Multiple Choice Questions: Reasoning Makes Large Language Models (LLMs) More Self-Confident, Especially When They are Wrongcited by 11
- 2025Advancing the Scientific Method with Large Language Models: From Hypothesis to Discoverycited by 7
- 2025Teaching Language Models to Faithfully Express their Uncertaintycited by 7
- 2026Reasoning Models Will Sometimes Lie About Their Reasoningcited by 6
- 2026Towards a Mechanistic Understanding of Large Reasoning Models: A Survey of Training, Inference, and Failurescited by 6
- 2026LLMs Should Express Uncertainty Explicitlycited by 6
- 2026Do explanations generalize across large reasoning models?cited by 5
- 2025Demystifying Scientific Problem-Solving in LLMs by Probing Knowledge and Reasoningcited by 4
- 2026Wired for Overconfidence: A Mechanistic Perspective on Inflated Verbalized Confidence in LLMscited by 4
Built on by
Citing papers per year: 2026: 2.
2 papers.
Similar, not cited
Close in wording, no citation either way
- 2026Claim-Level Confidence Calibration for Reliable Decision Making with Large Language Models
- 2026Closing the Confidence-Faithfulness Gap in Large Language Models
- 2025Trace Length is a Simple Uncertainty Signal in Reasoning Models
- 2026Identifying Influential N-grams in Confidence Calibration via Regression Analysis
- 2026Rethinking Uncertainty Evaluation in Large Language Models
- 2026Probability is Not Enough: Exploring and Counting Divergent Tokens for Reasoning Uncertainty Quantification in LLMs
- 2026Calibration Drift Under Reasoning: How Chain-of-Thought Budgets Induce Overconfidence in Large Language Models
- 2026Unsupervised Confidence Calibration for Reasoning LLMs from a Single Generation