mlsanityData to 14 Oct 2026

More Thinking, Less Seeing? Assessing Amplified Hallucination in Multimodal Reasoning Models

arXiv 2505.21523 ↗First posted 23 May 2025cs.CL cs.AI cs.CVcited by 86

On the map: LLM reasoning → Hallucination detection and mitigation

also in 1 more
Abstract (arXiv)

Test-time compute has empowered multimodal large language models to generate extended reasoning chains, yielding strong performance on tasks such as multimodal math reasoning. However, this improved reasoning ability often comes with increased hallucination: as generations become longer, models tend to drift away from image-grounded content and rely more heavily on language priors. Attention analysis shows that longer reasoning chains lead to reduced focus on visual inputs, which contributes to hallucination. To systematically study this phenomenon, we introduce RH-AUC, a metric that quantifies how a model's perception accuracy changes with reasoning length, allowing us to evaluate whether the model preserves visual grounding during reasoning. We also release RH-Bench, a diagnostic benchmark that spans a variety of multimodal tasks, designed to assess the trade-off between reasoning ability and hallucination. Our analysis reveals that (i) larger models typically achieve a better balance between reasoning and perception, and (ii) this balance is influenced more by the types and domains of training data than by its overall volume. These findings underscore the importance of evaluation frameworks that jointly consider both reasoning quality and perceptual fidelity.

Key facts. Cited by 86 papers in the map (data through 14 Oct 2026). Placed in Hallucination detection and mitigation, one of 17 directions in LLM reasoning. Builds on 47 of its 58 references that are in the map. First posted on arXiv 23 May 2025.

Builds on

47 of 58 references.

References · 47

  1. 2023MMBench: Is Your Multi-modal Model an All-around Player?cited by 2,141
  2. 2023MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGIcited by 2,080
  3. 2022Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answeringcited by 1,919
  4. 2023Evaluating Object Hallucination in Large Vision-Language Modelscited by 1,729
  5. 2023MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contextscited by 1,614
  6. 2024Are We on the Right Way for Evaluating Large Vision-Language Models?cited by 914
  7. 2024Eyes Wide Shut? Exploring the Visual Shortcomings of Multimodal LLMscited by 735
  8. 2023Aligning Large Multimodal Models with Factually Augmented RLHFcited by 712
  9. 2018Object Hallucination in Image Captioningcited by 669
  10. 2025Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Modelscited by 665
  11. 2023HallusionBench: An Advanced Diagnostic Suite for Entangled Language Hallucination and Visual Illusion in Large Vision-Language Modelscited by 583
  12. 2024Measuring Multimodal Mathematical Reasoning with MATH-Vision Datasetcited by 526
35 more
  1. 2025VLM-R1: A Stable and Generalizable R1-style Large Vision-Language Modelcited by 526
  2. 2025Visual-RFT: Visual Reinforcement Fine-Tuningcited by 513
  3. 2023OPERA: Alleviating Hallucination in Multi-Modal Large Language Models via Over-Trust Penalty and Retrospection-Allocationcited by 478
  4. 2024LLaVA-CoT: Let Vision Language Models Reason Step-by-Stepcited by 454
  5. 2023RLHF-V: Towards Trustworthy MLLMs via Behavior Alignment from Fine-grained Correctional Human Feedbackcited by 412
  6. 2025R1-Onevision: Advancing Generalized Multimodal Reasoning through Cross-Modal Formalizationcited by 370
  7. 2025Demystifying Long Chain-of-Thought Reasoning in LLMscited by 326
  8. 2025VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learningcited by 323
  9. 2025R1-VL: Learning to Reason with Multimodal Large Language Models via Step-wise Group Relative Policy Optimizationcited by 301
  10. 2024We-Math: Does Your Large Multimodal Model Achieve Human-like Mathematical Reasoning?cited by 286
  11. 2025LMM-R1: Empowering 3B LMMs with Strong Reasoning Abilities Through Two-Stage Rule-Based RLcited by 248
  12. 2023In-context Vectors: Making In Context Learning More Effective and Controllable Through Latent Space Steeringcited by 245
  13. 2025Seg-Zero: Reasoning-Chain Guided Segmentation via Cognitive Reinforcementcited by 239
  14. 2024Mulberry: Empowering MLLM with o1-like Reasoning and Reflection via Collective Monte Carlo Tree Searchcited by 175
  15. 2025PhysBench: Benchmarking and Enhancing Vision-Language Models for Physical World Understandingcited by 160
  16. 2025LlamaV-o1: Rethinking Step-by-step Visual Reasoning in LLMscited by 147
  17. 2025SoTA with Less: MCTS-Guided Sample Selection for Data-Efficient Visual Reasoning Self-Improvementcited by 145
  18. 2024Marco-o1: Towards Open Reasoning Models for Open-Ended Solutionscited by 144
  19. 2025VisualPRM: An Effective Process Reward Model for Multimodal Reasoningcited by 139
  20. 2024Improve Vision Language Model Chain-of-thought Reasoningcited by 136
  21. 2023Holistic Analysis of Hallucination in GPT-4V(ision): Bias and Interference Challengescited by 131
  22. 2024Insight-V: Exploring Long-Chain Visual Reasoning with Multimodal Large Language Modelscited by 116
  23. 2025Q-Insight: Understanding Image Quality via Visual Reinforcement Learningcited by 98
  24. 2024Self-Introspective Decoding: Alleviating Hallucinations for Large Vision-Language Modelscited by 94
  25. 2024LLaMA-Berry: Pairwise Optimization for O1-like Olympiad-Level Mathematical Reasoningcited by 91
  26. 2025Boosting the Generalization and Reasoning of Vision Language Models with Curriculum Reinforcement Learningcited by 71
  27. 2025RedStar: Does Scaling Long-CoT Data Unlock Better Slow-Reasoning Systems?cited by 60
  28. 2025GuardReasoner-VL: Safeguarding VLMs via Reinforced Reasoningcited by 42
  29. 2025ClearSight: Visual Signal Enhancement for Object Hallucination Mitigation in Multimodal Large language Modelscited by 40
  30. 2025Fast-Slow Thinking GRPO for Large Vision-Language Model Reasoningcited by 36
  31. 2024MM-MATH: Advancing Multimodal Math Evaluation with Process Evaluation and Fine-grained Classificationcited by 35
  32. 2025Automated Generation of Challenging Multiple-Choice Questions for Vision Language Model Evaluationcited by 30
  33. 2024PaCE: Parsimonious Concept Engineering for Large Language Modelscited by 19
  34. 2024MMEvalPro: Calibrating Multimodal Benchmarks Towards Trustworthy and Efficient Evaluationcited by 18
  35. 2025Content-Rich AIGC Video Quality Assessment via Intricate Text Alignment and Motion-Aware Consistencycited by 12

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 23; 2026: 63.

86 papers.

Sort:

By area:

  1. 2025Self-Rewarding Vision-Language Model via Reasoning Decompositioncited by 97
  2. 2025Reinforcement Learning Meets Large Language Models: A Survey of Advancements and Applications Across the LLM Lifecyclecited by 35
  3. 2025Open Vision Reasoner: Transferring Linguistic Cognitive Behavior for Visual Reasoningcited by 34
  4. 2025Reasoning Within the Mind: Dynamic Multimodal Interleaving in Latent Spacecited by 28
  5. 2025More Thought, Less Accuracy? On the Dual Nature of Reasoning in Vision-Language Modelscited by 27
  6. 2025A Comprehensive Survey on Trustworthiness in Reasoning with Large Language Modelscited by 25
  7. 2025Perceptual-Evidence Anchored Reinforced Learning for Multimodal Reasoningcited by 20
  8. 2025MVI-Bench: A Comprehensive Benchmark for Evaluating Robustness to Misleading Visual Inputs in LVLMscited by 15
  9. 2025Understanding Language Prior of LVLMs by Contrasting Chain-of-Embeddingcited by 14
  10. 2025Think Twice to See More: Iterative Visual Reasoning in Medical VLMscited by 13
  11. 2026Thinking in Uncertainty: Mitigating Hallucinations in MLRMs with Latent Entropy-Aware Decodingcited by 12
  12. 2025Test-Time Scaling in Reasoning Models Is Not Effective for Knowledge-Intensive Tasks Yetcited by 11
38 more
  1. 2026SceneAlign: Aligning Multimodal Reasoning to Scene Graphs in Complex Visual Scenescited by 11
  2. 2026On Robustness and Chain-of-Thought Consistency of RL-Finetuned VLMscited by 11
  3. 2025Evaluating Reasoning Faithfulness in Medical Vision-Language Models using Multimodal Perturbationscited by 9
  4. 2026Seeing with You: Perception-Reasoning Coevolution for Multimodal Reasoningcited by 9
  5. 2026H-OPD: Confidence Aware Heterogeneous Multi-Teacher Multimodal On-policy Distillationcited by 9
  6. 2025RadAgents: Multimodal Agentic Reasoning for Chest X-ray Interpretation with Radiologist-like Workflowscited by 8
  7. 2026From Narrow to Panoramic Vision: Attention-Guided Cold-Start Reshapes Multimodal Reasoningcited by 8
  8. 2026Beyond SFT-to-RL: Pre-alignment via Black-Box On-Policy Distillation for Multimodal RLcited by 8
  9. 2025Chart-RVR: Reinforcement Learning with Verifiable Rewards for Explainable Chart Reasoningcited by 7
  10. 2026WorldMemArena: Evaluating Multimodal Agent Memory Through Action-World Interactioncited by 7
  11. 2025Journey Before Destination: On the importance of Visual Faithfulness in Slow Thinkingcited by 6
  12. 2026From Seeing to Thinking: Decoupling Perception and Reasoning Improves Post-Training of Vision-Language Modelscited by 6
  13. 2025Pay More Attention To Audio: Mitigating Imbalance of Cross-Modal Attention in Large Audio Language Modelscited by 5
  14. 2026Seeing Through the Chain: Mitigate Hallucination in Multimodal Reasoning Models via CoT Compression and Contrastive Preference Optimizationcited by 5
  15. 2026GraphThinker: Reinforcing Temporally Grounded Video Reasoning with Event Graph Thinkingcited by 5
  16. 2026Deeper Thought, Weaker Aim: Understanding and Mitigating Perceptual Impairment during Reasoning in Multimodal Large Language Modelscited by 5
  17. 2026HopChain: Multi-Hop Data Synthesis for Generalizable Vision-Language Reasoningcited by 5
  18. 2026Large Vision-Language Models Get Lost in Attentioncited by 5
  19. 2025Reallocating Attention Across Layers to Reduce Multimodal Hallucinationcited by 4
  20. 2026TabSieve: Explicit In-Table Evidence Selection for Tabular Predictioncited by 4
  21. 2026What does RL improve for Visual Reasoning? A Frankenstein-Style Analysiscited by 4
  22. 2026Efficient Reasoning with Balanced Thinkingcited by 4
  23. 2026Improving Vision-language Models with Perception-centric Process Reward Modelscited by 4
  24. 2026Attend to Evidence: Evidence-Anchored Spatial Attention Supervision for Multimodal RLVRcited by 4
  25. 2025Unleashing Perception-Time Scaling to Multimodal Reasoning Modelscited by 3
  26. 2025Learning to Think Fast and Slow for Visual Language Modelscited by 3
  27. 2026Ground What You See: Hallucination-Resistant MLLMs via Caption Feedback, Diversity-Aware Sampling, and Conflict Regularizationcited by 3
  28. 2026Understanding and Mitigating Hallucinations in Multimodal Chain-of-Thought Modelscited by 3
  29. 2025Video Parallel Scaling: Aggregating Diverse Frame Subsets for VideoLLMscited by 2
  30. 2025Decouple to Generalize: Context-First Self-Evolving Learning for Data-Scarce Vision-Language Reasoningcited by 2
  31. 2026SAVAA: Mitigating Hallucinations in LVLMs via Step-wise Adaptive Visual Attention Amplificationcited by 2
  32. 2026StAR: Segment Anything Reasonercited by 2
  33. 2026Beyond Perceptual Shortcuts: Causal-Inspired Debiasing Optimization for Generalizable Video Reasoning in Lightweight MLLMscited by 2
  34. 2026Hallucination as Exploit: Evidence-Carrying Multimodal Agentscited by 2
  35. 2026HalluWorld: A Controlled Benchmark for Hallucination via Reference World Modelscited by 2
  36. 2026EvoEmbedding: Evolvable Representations for Long-Context Retrieval and Agentic Memorycited by 2
  37. 2026No Place to Hide: Benchmarking Video Hallucination with Background-Controlled Pairscited by 2
  38. 2026MM-THEBench: Do Reasoning MLLMs Think Reasonably?cited by 1

50 most cited of 86 All 86 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2024Thinking Before Looking: Improving Multimodal LLM Reasoning via Mitigating Visual Hallucination
  2. 2026HIVE: Understanding Post-Hallucination Reasoning in Vision Language Models
  3. 2025MIRAGE: Assessing Hallucination in Multimodal Reasoning Chains of MLLM
  4. 2025Detection and Mitigation of Hallucination in Large Reasoning Models: A Mechanistic Perspective
  5. 2025Critique Before Thinking: Mitigating Hallucination through Rationale-Augmented Instruction Tuning
  6. 2025Auditing Meta-Cognitive Hallucinations in Reasoning Large Language Models
  7. 2026What Do Hallucinations Reveal About Multimodal Reasoning? Diagnosing Visual Grounding Failures via Contrastive Decoding Probes
  8. 2024Combating Multimodal LLM Hallucination via Bottom-Up Holistic Reasoning