mlsanityData to 7 Oct 2026

Instinct vs. Reflection: Unifying Token and Verbalized Confidence in Multimodal Large Models

arXiv 2604.17274 ↗First posted 19 Apr 2026cs.CV cs.AIcited by 3

On the map: Multimodal LLMs / vision-language models → MLLM training and visual instruction tuning

Abstract (arXiv)

Multimodal Large Language Models (MLLMs) have demonstrated exceptional capabilities in various perception and reasoning tasks. Despite this success, ensuring their reliability in practical deployment necessitates robust confidence estimation. Prior works have predominantly focused on text-only LLMs, often relying on computationally expensive self-consistency sampling. In this paper, we extend this to multimodal settings and conduct a comprehensive evaluation of MLLMs' response confidence estimation. Our analysis reveals a significant instinct-reflection misalignment: the model's implicit token-level support frequently diverges from its verbal self-assessment confidence. To address this misalignment, we propose a monotone confidence fusion framework to merge dual-channel signals and cross-channel consistency to estimate correctness. Subsequently, an order-preserving mean alignment step is applied to correct global bias, which improves calibration while preserving the risk-coverage trade-off for selective prediction. Experiments on diverse open-source and closed-source MLLMs show that our method consistently yields more reliable confidence estimates and improves both calibration and failure prediction. Code will be available at https://github.com/Yunkaidang/Instinct-vs.-Reflection.

Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Placed in MLLM training and visual instruction tuning, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 45 of its 58 references that are in the map. First posted on arXiv 19 Apr 2026.

Builds on

45 of 58 references.

Closest ancestors · 6

Same direction, newest first

  1. 2024Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling6 Dec
  2. 2024Explainable and Interpretable Multimodal Large Language Models: A Comprehensive Survey3 Dec
  3. 2024Exploring Response Uncertainty in MLLMs: An Empirical Evaluation under Misleading Scenarios5 Nov
  4. 2024Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution18 Sep
  5. 2024MiniCPM-V: A GPT-4V Level MLLM on Your Phone3 Aug
  6. 2023MMBench: Is Your Multi-modal Model an All-around Player?12 Jul

Other references · 39

  1. 2023GPT-4 Technical Reportcited by 18,132
  2. 2025Qwen2.5-VL Technical Reportcited by 5,418
  3. 2017On Calibration of Modern Neural Networkscited by 4,860
  4. 2024GPT-4o System Cardcited by 4,467
  5. 2022Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answeringcited by 1,881
  6. 2025InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Modelscited by 1,769
  7. 2023MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contextscited by 1,582
  8. 2025InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiencycited by 1,419
  9. 2019Can You Trust Your Model's Uncertainty? Evaluating Predictive Uncertainty Under Dataset Shiftcited by 1,274
  10. 2023Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMscited by 946
  11. 2024Are We on the Right Way for Evaluating Large Vision-Language Models?cited by 893
  12. 2016A Diagram Is Worth A Dozen Imagescited by 793
27 more
  1. 2023Just Ask for Calibration: Strategies for Eliciting Calibrated Confidence Scores from Language Models Fine-Tuned with Human Feedbackcited by 788
  2. 2024DeepSeek-VL: Towards Real-World Vision-Language Understandingcited by 718
  3. 2024Phi-4 Technical Reportcited by 675
  4. 2024MMMU-Pro: A More Robust Multi-discipline Multimodal Understanding Benchmarkcited by 494
  5. 2019Deep Ensembles: A Loss Landscape Perspectivecited by 438
  6. 2020Simple and Principled Uncertainty Estimation with Deterministic Deep Learning via Distance Awarenesscited by 389
  7. 2019Beyond temperature scaling: Obtaining well-calibrated multiclass probabilities with Dirichlet calibrationcited by 359
  8. 2023A Survey of Confidence Estimation and Calibration in Large Language Modelscited by 253
  9. 2023Shifting Attention to Relevance: Towards the Predictive Uncertainty Quantification of Free-Form Large Language Modelscited by 147
  10. 2025Confidence Improves Self-Consistency in LLMscited by 111
  11. 2024Calibrating Language Models with Adaptive Temperature Scalingcited by 57
  12. 2024A Survey of Uncertainty Estimation in LLMs: Theory Meets Practicecited by 54
  13. 2023Knowing What LLMs DO NOT Know: A Simple Yet Effective Self-Detection Methodcited by 51
  14. 2024Overconfidence is Key: Verbalized Uncertainty Evaluation in Large Language and Vision-Language Modelscited by 51
  15. 2024Think Twice Before Trusting: Self-Detection for Large Language Models through Comprehensive Answer Reflectioncited by 40
  16. 2024Uncertainty in Language Models: Assessment through Rank-Calibrationcited by 40
  17. 2024Thermometer: Towards Universal Calibration for Large Language Modelscited by 38
  18. 2024Confidence Under the Hood: An Investigation into the Confidence-Probability Alignment in Large Language Modelscited by 34
  19. 2025Optimizing Temperature for Language Models with Multi-Sample Inferencecited by 33
  20. 2025Calibrating LLM Confidence by Probing Perturbed Representation Stabilitycited by 23
  21. 2024Unveiling the Tapestry of Consistency in Large Vision-Language Modelscited by 22
  22. 2024An Empirical Study Into What Matters for Calibrating Vision-Language Modelscited by 20
  23. 2025Seeing is Believing, but How Much? A Comprehensive Analysis of Verbalized Calibration in Vision-Language Modelscited by 16
  24. 2025Do Large Language Models Know How Much They Know?cited by 11
  25. 2025Revisiting Self-Consistency from Dynamic Distributional Alignment Perspective on Answer Aggregationcited by 11
  26. 2025Object-Level Verbalized Confidence Calibration in Vision-Language Models via Semantic Perturbationcited by 11
  27. 2024Multi-Perspective Consistency Enhances Confidence Estimation in Large Language Modelscited by 4

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 3.

3 papers.

  1. 2026Calibrated Triage, Not Autonomy: Confidence Estimation for Medical Vision-Language Modelscited by 2
  2. 2026Ranking-Aware Calibration for Reliable Multimodal Reinforcement Learningcited by 1
  3. 2026Uncertainty-Aware Decision Making in Multimodal Large Language Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Linking Perception, Confidence and Accuracy in MLLMs
  2. 2025Direct Confidence Alignment: Aligning Verbalized Confidence with Internal Confidence In Large Language Models
  3. 2026ConfidenceBench: Evaluating Confidence Calibration in Large Language Models
  4. 2024Calibrating the Confidence of Large Language Models by Eliciting Fidelity
  5. 2026Uncertainty Quantification for Multimodal Large Language Models with Incoherence-adjusted Semantic Volume
  6. 2026Rethinking Uncertainty Evaluation in Large Language Models
  7. 2026Closing the Confidence-Faithfulness Gap in Large Language Models
  8. 2024Unveiling Uncertainty: A Deep Dive into Calibration and Performance of Multimodal Large Language Models