Reasoning Verification

Follow this topic Ask about this topic

Reasoning verification studies how to detect flawed steps, assess proof validity, and determine whether answers follow from evidence rather than merely matching an answer key. Approaches include process reward models and step classifiers, self- or multi-agent checking, symbolic and formal proof checkers, and evidence-grounded tools for tasks such as mathematics, code, and multi-hop fact verification. Benchmarks increasingly test first-error localization, robustness to valid but non-canonical solutions, and whether a verifier depends on the evidence it is given. A central challenge is balancing reliable detection against false rejection, reward hacking, and the token or human-review costs of stronger checks.

Latest papers

57 papers9 papers in the last four weeks, up 80% on the four weeks before. 0.1% of all new papers.

Stepwise Reasoning Verification: Don't Generate, Classify!
Oct 18, 2026Yifei Shi, Michalis Korakakis, Rami Aly +3Reasoning Verification Language Model Reasoning Evaluation
University of Cambridge · Alan Turing Institute · University of Edinburgh
Verifying Human-Written Reasoning References
Oct 17, 2026Ali Ansari, Farhana Ferdousi LizaReasoning Verification Language Model Reasoning Evaluation
School of Computing Sciences, University of East Anglia
AlphaDiana: A Measurement System for Agents in Verifiable Reasoning Tasks
Oct 14, 2026Zhanke Zhou, Zongze Li, Weikai Huang +16LLM Agent Evaluation LLM Agent Harnesses
TMLR Group, Hong Kong Baptist University · University of Arizona · Zuse Institute Berlin +3
Beyond Answer Voting: Fine-Grained Correction of LLM Reasoning through Graph Alignment across Sampled Traces
Oct 13, 2026Haotian Liu, Yingjian Chen, Xiaohang Xu +4LLM Self-Correction Reasoning Verification
School of Informatics, Xiamen University · The University of Tokyo · National Institute of Informatics (NII)
One-Sided Verification Under Limited Expert Review
Oct 13, 2026Fan Nie, Haotian Ye, Shurui Liu +3Reasoning Verification Human-in-the-Loop Evaluation
Stanford University
Attention-Graph Curvature for the Verification of Mathematical Reasoning
Oct 12, 2026Kanghyeon Kim, Johannes F. Lutzeyer, Yanzhu GuoLanguage Model Error Detection Reasoning Verification
Télécom Paris, IPP · LIX, CNRS, École Polytechnique, IPP
Can Language Models Learn to Reject Their Own Bad Reasoning Steps?
Oct 5, 2026Siheng Xiong, Xiaoze Liu, Yiqiao Jin +2LLM Self-Correction Language Model Decoding
Georgia Institute of Technology · Purdue University
Hallucination Across the Reasoning Lifecycle: Interface Visibility, Causal Evidence, and Release Control in Large Reasoning Models
Oct 4, 2026Zhe Yu, Mohan Li, Lei Yu +6LLM Hallucination Mitigation Hallucination in Language Models
Zhejiang University, China · Binjiang Institute of Zhejiang University, China · Guangzhou University, China +5
Detecting Inconsistencies in Model Specifications with LLM-as-Verifier Reasoning
Oct 1, 2026Zichen Xie, Mrigank Pawagi, Lize Shao +2LLM Alignment LLM Auditing
University of Virginia · University of Pennsylvania · The University of Texas at Austin
The Error You See Is Not the Error You Made: Progression-aware Reasoning Origin for Reasoning Error Localization
Sep 27, 2026Yiguo Wang, Ziyuan Yang, Yi Zou +3Language Model Error Detection Reasoning Verification
Xiaopeng Honors College, Nanchang Hangkong University, China · School of Cyber Science and Engineering, Sichuan University, China · School of Software, Nanchang Hangkong University, China +1
BudgetVerify: Budget-Tiered Verification for Financial QA
Sep 27, 2026Janet Jenq, Hongda ShenLLM Answer Verification Reasoning Verification
PitchBook
Route-Verify-Vote: Procedure-Conditioned Self-Consistency for Mixed-Domain Reasoning
Sep 23, 2026Xinchen XiaoCommonsense Reasoning Self-Consistency in Language Models
Xinjiang University
SoK: Formal Methods for Fact-Checking and Information Integrity
Sep 19, 2026Nikolaos Kekatos, Theodoros Nestoridis, Charalampos Bratsas +5Automated Fact-Checking Reasoning Verification
Clone Systems, Cyprus · International Hellenic University, Greece · School of Journalism and Mass Communications, Aristotle University of Thessaloniki, Greece +2
Deposon: An Auditable, Conservation-Guaranteed, Game-Theoretically Tested Scattering Layer over LLM Reasoning Paths
Sep 8, 2026Qihao YuanLLM Reasoning with Graphs LLM Auditing
School of Chemistry and Life Resources, Renmin University of China
Beyond the Answer Key: Robustness Evaluation of Large Language Models for Step-Level Mathematical Verification
Aug 28, 2026Fateme Mazdarani, Carlos ToxtliLLM Evaluation LLM Answer Verification
School of Computing Clemson University Clemson, SC, USA
Does the Proof Prove It That Way? Faithful Formalization of Elements Proofs
Aug 15, 2026Tadd Mao, Tianjun Zhong, Dhruva Arekar +5Automated Theorem Proving Reasoning Verification
University of Toronto · Johns Hopkins University · University of Pennsylvania
ReflectFact: Self-Reflective Agents for Improving Comprehension and Reasoning in Multi-Hop Fact Verification
Aug 13, 2026Runze Zhao, Zixin Tang, Xiaoshuai Hao +4Self-Reflection in Language Model Agents Automated Fact-Checking
Institute of Information Engineering, Chinese Academy of Sciences · 1Zhongguancun Laboratory · 3Xiaomi EV
Claim-Level Reliability Assessment for Efficient Test-Time Reasoning
Aug 12, 2026Sen Xu, Wei Wang, Shixi Liu +5Test-Time Scaling LLM Reliability
Sina Weibo Inc.
A Human Audit of OpenAIs AI-Generated Mathematical Proofs
Aug 3, 2026Mikołaj Sienicki, Krzysztof SienickiLLM Auditing Human Oversight of AI
Polish-Japanese Academy of Information Technology ul. Koszykowa 86, 02-008 Warsaw, Poland, European Union · Chair of Theoretical Physics of Naturally Intelligent Systems (NIS)1 ul. Lipowa 2 / Topolowa 19, 05-807 Podkowa Leśna, Poland, European Union
Answer-Conditioned Chains of Thought Degrade Verifiable-Reasoning Distillation in Large Language Models
Jul 16, 2026Jungseob Lee, Seungyoon Lee, Suhyune Son +4CoT Distillation Language Model Distillation
1Korea University · 2Zoom Communications · 3Yonsei University
Evidence-Grounded Verified Agentic Reasoning: A Path Toward Eliminating LLM Hallucination in Empirical Inference via Tool-Attested Kernel Proofs
Jul 14, 2026Junyu RenLLM Hallucination Mitigation Evidence-Grounded Reasoning
Committee on Computational and Applied Mathematics, University of Chicago, Chicago, IL, USA.
Forethought: Verifiable Reasoning from Neurosymbolic Primitive Programming
Jul 5, 2026Vishvesh Bhat, Jay Vaghasiya, Emmanuel Anaya GonzalezSymbolic Reasoning Efficient Language Model Reasoning
CoreThink AI · UC San Diego
CheckRLM: Effective Knowledge-Thought Coherence Checking in Retrieval-Augmented Reasoning
Jul 2, 2026Dingling Xu, Ruobing Wang, Qingfei Zhao +8Retrieval-Augmented Reasoning LLM Reliability
School of Artificial Intelligence, Beijing Normal University, Beijing 100875, China · Beijing Key Laboratory of Artificial Intelligence for Education, Beijing 100875, China · Institute of Information Engineering, Chinese Academy of Sciences +3
Theoria: Rewrite-Acceptability Verification over Informal Reasoning States
Jul 1, 2026Michael Saldivar, Ben SlivinskiLLM Answer Verification LLM Auditing
Independent Researchers
Think Again or Think Longer? Selective Verification for Budget-Aware Reasoning
Jun 18, 2026Sajib Acharjee Dip, Dawei Zhou, Liqing ZhangLLM Inference Efficiency Cost-Aware Inference
Department of Computer Science, Virginia Tech · Fralin Biomedical Research Institute, Virginia Tech · FBRI Cancer Research Center, Washington, DC
Scaling LLM Reasoning from Minimal Labels: A Semi-Supervised Framework with a Lightweight Verifier
Jun 15, 2026Keizo Kato, Chenhui Chu, Yugo Murawaki +1Pseudo-Labeling LLM Reasoning
Fujitsu Limited, Kawasaki, Japan · Kyoto University, Kyoto, Japan · National Institute of Informatics, Tokyo, Japan
VibeThinker-3B: Exploring the Frontier of Verifiable Reasoning in Small Language Models
Jun 15, 2026Sen Xu, Shixi Liu, Wei Wang +6Small Language Models Reasoning Verification
Sina Weibo Inc.
Odds Law: The Decomposition Algebra On How Intelligence Organizes Itself to Solve Difficult Problems Reliably
Jun 14, 2026Hidayet AksuAgent Reliability Ensemble Learning
VeriGeo: Controllable Geometry Question Generation with Numerical and Analytical Verification
Jun 12, 2026Xiaoxian Duan, Zequn Liu, Yingce XiaSynthetic Data Generation Multimodal Reasoning
Institute of Automation, Chinese Academy of Sciences, Beijing, China · Zhongguancun Academy, Beijing, China
Evaluating Research-Level Math Proofs via Strict Step-Level Verification
Jun 9, 2026Yifeng SunLLM Answer Verification Automated Theorem Proving
Independent Researcher

Oct 18, 2026J/K move · Enter open · S save

Stepwise Reasoning Verification: Don't Generate, Classify!

Yifei Shi, Michalis Korakakis, Rami Aly, Eric Chamoun+2

University of Cambridge · Alan Turing Institute · University of Edinburgh

Reasoning Verification Language Model Reasoning Evaluation Decoder-Only Language Models

Open paper AskPDF arXiv

Verifying the reasoning of large language models is increasingly important as they are applied to complex mathematical and scientific problems. Existing work focuses on large decoder models and reports mixed results for generative and discriminative verifiers, leaving unclear which paradigm is more effective and whether large models are necessary. In this work, we cast stepwise verification as a natural language inference task: the problem and partial solution prefix form the premise, while the next reasoning step serves as the hypothesis to be classified as entailed or non-entailed. This formulation enables error localisation at individual reasoning steps and direct comparison of encoder classification, decoder classification, and decoder verdict generation across model scales and annotation budgets. Across five reasoning datasets, a 0.15B encoder classifier outperforms most generative verifiers with up to 4B parameters. On identical decoder backbones, classification outperforms generation from 0.5B to 4B parameters and matches it on the strongest 7B backbone---reaching comparable performance with roughly one tenth as many annotated problems. Our findings establish classification as a parameter- and data-efficient approach to stepwise verification and show that strong verification does not require large generative decoders.