mlsanityData to 7 Oct 2026

AbstentionBench: Reasoning LLMs Fail on Unanswerable Questions

arXiv 2506.09038 ↗First posted 10 Jun 2025cs.AIcited by 129

On the map: LLM reasoning → Reasoning benchmarks and evaluation

Abstract (arXiv)

For Large Language Models (LLMs) to be reliably deployed in both everyday and high-stakes domains, knowing when not to answer is equally critical as answering correctly. Real-world user queries, which can be underspecified, ill-posed, or fundamentally unanswerable, require LLMs to reason about uncertainty and selectively abstain -- i.e., refuse to answer definitively. However, abstention remains understudied, without a systematic evaluation framework for modern LLMs. In this work, we introduce AbstentionBench, a large-scale benchmark for holistically evaluating abstention across 20 diverse datasets, including questions with unknown answers, underspecification, false premises, subjective interpretations, and outdated information. Evaluating 20 frontier LLMs reveals abstention is an unsolved problem, and one where scaling models is of little use. While recent reasoning LLMs have shown impressive results in complex problem solving, surprisingly, we find that reasoning fine-tuning degrades abstention (by 24% on average), even for math and science domains on which reasoning models are explicitly trained. We find that while a carefully crafted system prompt can boost abstention in practice, it does not resolve models' fundamental inability to reason about uncertainty. We release AbstentionBench to foster research into advancing LLM reliability.

Key facts. Cited by 129 papers in the map (data through 7 Oct 2026). Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning. Builds on 75 of its 86 references that are in the map. First posted on arXiv 10 Jun 2025.

Builds on

75 of 86 references.

Closest ancestors · 1

Same direction, newest first

  1. 2023WorldSense: A Synthetic Benchmark for Grounded Reasoning in Large Language Models27 Nov

Other references · 74

  1. 2022Training language models to follow instructions with human feedbackcited by 15,712
  2. 2017Proximal Policy Optimization Algorithmscited by 14,441
  3. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  4. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  5. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,393
  6. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
  7. 2020Measuring Massive Multitask Language Understandingcited by 7,768
  8. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  9. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,767
  10. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,203
  11. 2023Let's Verify Step by Stepcited by 3,598
  12. 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,748
38 more
  1. 2021SimCSE: Simple Contrastive Learning of Sentence Embeddingscited by 2,168
  2. 2022Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language modelscited by 1,970
  3. 2022Language Models (Mostly) Know What They Knowcited by 1,783
  4. 2023A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questionscited by 1,582
  5. 2018Know What You Don't Know: Unanswerable Questions for SQuADcited by 1,538
  6. 2019Do ImageNet Classifiers Generalize to ImageNet?cited by 1,507
  7. 2022Solving Quantitative Reasoning Problems with Language Modelscited by 1,359
  8. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 1,330
  9. 2025s1: Simple test-time scalingcited by 1,311
  10. 2022Program of Thoughts Prompting: Disentangling Computation from Reasoning for Numerical Reasoning Taskscited by 1,057
  11. 2024Qwen2.5-Math Technical Report: Toward Mathematical Expert Model via Self-Improvementcited by 1,057
  12. 2023Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMscited by 946
  13. 2022STaR: Bootstrapping Reasoning With Reasoningcited by 895
  14. 2021MuSiQue: Multihop Questions via Single-hop Question Compositioncited by 893
  15. 2023Large Language Models Can Be Easily Distracted by Irrelevant Contextcited by 820
  16. 2023Just Ask for Calibration: Strategies for Eliciting Calibrated Confidence Scores from Language Models Fine-Tuned with Human Feedbackcited by 788
  17. 2022Teaching Models to Express Their Uncertainty in Wordscited by 731
  18. 2023MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Modelscited by 724
  19. 2021BBQ: A Hand-Built Bias Benchmark for Question Answeringcited by 683
  20. 2023WizardMath: Empowering Mathematical Reasoning for Large Language Models via Reinforced Evol-Instructcited by 618
  21. 2021A Dataset of Information-Seeking Questions and Answers Anchored in Research Paperscited by 475
  22. 2025Reasoning Models Don't Always Say What They Thinkcited by 388
  23. 2024Hallucination is Inevitable: An Innate Limitation of Large Language Modelscited by 329
  24. 2023LegalBench: A Collaboratively Built Benchmark for Measuring Legal Reasoning in Large Language Modelscited by 328
  25. 2023FreshLLMs: Refreshing Large Language Models with Search Engine Augmentationcited by 308
  26. 2025MathArena: Evaluating LLMs on Uncontaminated Math Competitionscited by 275
  27. 2023Do Large Language Models Know What They Don't Know?cited by 238
  28. 2025Dynamic Early Exit in Reasoning Modelscited by 217
  29. 2023Evaluating the Moral Beliefs Encoded in LLMscited by 211
  30. 2024Judging the Judges: Evaluating Alignment and Vulnerabilities in LLMs-as-Judgescited by 197
  31. 2024Don't Hallucinate, Abstain: Identifying LLM Knowledge Gaps via Multi-LLM Collaborationcited by 176
  32. 2021SituatedQA: Incorporating Extra-Linguistic Contexts into QAcited by 145
  33. 2024MediQ: Question-Asking LLMs and a Benchmark for Reliable Interactive Clinical Reasoningcited by 145
  34. 2024Know Your Limits: A Survey of Abstention in Large Language Modelscited by 116
  35. 2024Benchmarking Uncertainty Quantification Methods for Large Language Models with LM-Polygraphcited by 105
  36. 2021Reduced, Reused and Recycled: The Life of a Dataset in Machine Learning Researchcited by 100
  37. 2024The Art of Saying No: Contextual Noncompliance in Language Modelscited by 100
  38. 2024Large Language Models Must Be Taught to Know What They Don't Knowcited by 99

50 most cited of 74 All 74 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 23; 2026: 106.

129 papers.

Sort:

By area:

  1. 2025Beyond Binary Rewards: Training LMs to Reason About Their Uncertaintycited by 107
  2. 2025OptimalThinkingBench: Evaluating Over and Underthinking in LLMscited by 35
  3. 2025BrokenMath: A Benchmark for Sycophancy in Theorem Proving with LLMscited by 23
  4. 2025A Comprehensive Survey on Trustworthiness in Reasoning with Large Language Modelscited by 22
  5. 2025Trace Length is a Simple Uncertainty Signal in Reasoning Modelscited by 16
  6. 2026The Confidence Dichotomy: Analyzing and Mitigating Miscalibration in Tool-Use Agentscited by 14
  7. 2026Knowing When to Abstain: Medical LLMs Under Clinical Uncertaintycited by 13
  8. 2026Decoupling Reasoning and Confidence: Resurrecting Calibration in Reinforcement Learning from Verifiable Rewardscited by 13
  9. 2025Can LLMs Predict Their Own Failures? Self-Awareness via Internal Circuitscited by 12
  10. 2025L-MARS: Legal Multi-Agent System with Agentic Search and Citation-Faithfulness Auditcited by 11
  11. 2025Position: The Hidden Costs and Measurement Gaps of Reinforcement Learning with Verifiable Rewardscited by 11
  12. 2025Why Do Multilingual Reasoning Gaps Emerge in Reasoning Language Models?cited by 11
38 more
  1. 2026Abstain-R1: Calibrated Abstention and Post-Refusal Clarification via Verifiable RLcited by 11
  2. 2026Over-Searching in Search-Augmented Large Language Modelscited by 10
  3. 2025When Robots Should Say "I Don't Know": Benchmarking Abstention in Embodied Question Answeringcited by 9
  4. 2026RealFin: How Well Do LLMs Reason About Finance When Users Leave Things Unsaid?cited by 9
  5. 2026Causal Evidence that Language Models use Confidence to Drive Behaviorcited by 9
  6. 2026Agentic Abstention: Do Agents Know When to Stop Instead of Act?cited by 9
  7. 2025RefusalBench: Generative Evaluation of Selective Refusal in Grounded Language Modelscited by 8
  8. 2026Generate, Filter, Control, Replay: A Comprehensive Survey of Rollout Strategies for LLM Reinforcement Learningcited by 8
  9. 2026AgentAbstain: Do LLM Agents Know When Not to Act?cited by 8
  10. 2026HalluHard: A Hard Multi-Turn Hallucination Benchmarkcited by 7
  11. 2026Are LLM Decisions Faithful to Verbal Confidence?cited by 6
  12. 2026Reaching Beyond the Mode: RL for Distributional Reasoning in Language Modelscited by 6
  13. 2025Beyond Passive Critical Thinking: Fostering Proactive Questioning to Enhance Human-AI Collaborationcited by 5
  14. 2025KnowGuard: Knowledge-Driven Abstention for Multi-Round Clinical Reasoningcited by 5
  15. 2026When Silence Is Golden: Can LLMs Learn to Abstain in Temporal QA and Beyond?cited by 5
  16. 2026BAS: A Decision-Theoretic Approach to Evaluating Large Language Model Confidencecited by 5
  17. 2026TRIAGE: Evaluating Prospective Metacognitive Control in LLMs under Resource Constraintscited by 5
  18. 2025LLM Abstention Can Be a Prompt Artifact, in Addition to Genuine Uncertaintycited by 4
  19. 2025OffTopicEval: When Large Language Models Enter the Wrong Chat, Almost Always!cited by 4
  20. 2025Personalizing Agent Privacy Decisions via Logical Entailmentcited by 4
  21. 2026Statistical Early Stopping for Reasoning Modelscited by 4
  22. 2026Hallucinations Undermine Trust; Metacognition is a Way Forwardcited by 4
  23. 2026Coding Agents Don't Know When to Actcited by 4
  24. 2026The Yes-Man Syndrome: Benchmarking Abstention in Embodied Robotic Agentscited by 4
  25. 2026Just Ask Jev: Reinforcement Learning for Calibrated Decisions as a Zero-Shot Detector of AI Alignment Failurescited by 4
  26. 2025Answering the Unanswerable Is to Err Knowingly: Analyzing and Mitigating Abstention Failures in Large Reasoning Modelscited by 3
  27. 2025Stop Before You Fail: Operational Capability Boundaries for Mitigating Unproductive Reasoning in Large Reasoning Modelscited by 3
  28. 2025Honesty over Accuracy: Trustworthy Language Models through Reinforced Hesitationcited by 3
  29. 2025Knowing What's Missing: Assessing Information Sufficiency in Question Answeringcited by 3
  30. 2026ClaimDB: A Fact Verification Benchmark over Large Structured Datacited by 3
  31. 2026VLM-UQBench: A Benchmark for Modality-Specific and Cross-Modality Uncertainties in Vision Language Modelscited by 3
  32. 2026Confidence Before Answering: A Paradigm Shift for Efficient LLM Uncertainty Estimationcited by 3
  33. 2026Not All Proofs Are Equal: Evaluating LLM Proof Quality Beyond Correctnesscited by 3
  34. 2026Quantifying and Mitigating Premature Closure in Frontier LLMscited by 3
  35. 2026Aligning LLMs with Human Uncertainty: A Beta-Bernoulli Calibrator for LLM Forecastingcited by 3
  36. 2026TRAPSBench: Vision-Language Models Encode but Fail to Express Epistemic Restraintcited by 3
  37. 2025Toward Honest Language Models for Deductive Reasoningcited by 2
  38. 2026Unsupervised Confidence Calibration for Reasoning LLMs from a Single Generationcited by 2

50 most cited of 129 All 129 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2024LogicBench: Towards Systematic Evaluation of Logical Reasoning Ability of Large Language Models
  2. 2025What Defines Good Reasoning in LLMs? Dissecting Reasoning Steps with Multi-Aspect Evaluation
  3. 2025QuestBench: Can LLMs ask the right question to acquire information in reasoning tasks?
  4. 2024Critical-Questions-of-Thought: Steering LLM reasoning with Argumentative Querying
  5. 2026Knowing When Not to Answer: Abstention-Aware Scientific Reasoning
  6. 2025ReasoningWeekly: A General Knowledge and Verbal Reasoning Challenge for Large Language Models
  7. 2023Are Large Language Models Really Good Logical Reasoners? A Comprehensive Evaluation and Beyond
  8. 2025BeyondBench: Contamination-Resistant Evaluation of Reasoning in Language Models