mlsanityData to 14 Oct 2026

VLM Judges Can Rank but Cannot Score: Task-Dependent Uncertainty in Multimodal Evaluation

arXiv 2604.25235 ↗First posted 28 Apr 2026cs.LG cs.CL cs.CVcited by 8

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

Abstract (arXiv)

Vision-language models (VLMs) are increasingly used as automated judges for multimodal systems, yet their scores provide no indication of reliability. We study this problem through conformal prediction, a distribution-free framework that converts a judge's point score into a calibrated prediction interval using only score-token log-probabilities, with no retraining. We present the first systematic analysis of conformal prediction for VLM-as-a-Judge across 3 judges and 14 visual task categories. Our results show that evaluation uncertainty is strongly task-dependent: intervals cover ~40% of the score range for aesthetics and natural images but expand to ~70% for chart and mathematical reasoning, yielding a quantitative reliability map for multimodal evaluation. We further identify a failure mode not captured by standard evaluation metrics, ranking-scoring decoupling, where judges achieve high ranking correlation while producing wide, uninformative intervals, correctly ordering responses but failing to assign reliable absolute scores. Finally, we show that interval width is driven primarily by task difficulty and annotation quality, i.e., the same judge and method yield 4.5x narrower intervals on a clean, multi-annotator captioning benchmark. Code: https://github.com/divake/VLM-Judge-Uncertainty

Key facts. Cited by 8 papers in the map (data through 14 Oct 2026). Placed in MLLM training and visual instruction tuning, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 35 of its 40 references that are in the map. First posted on arXiv 28 Apr 2026.

Builds on

35 of 40 references.

Closest ancestors · 2

Same direction, newest first

  1. 2024LLaVA-Critic: Learning to Evaluate Multimodal Models3 Oct
  2. 2024MLLM-as-a-Judge: Assessing Multimodal LLM-as-a-Judge with Vision-Language Benchmark7 Feb

Other references · 33

  1. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  2. 2023G-Eval: NLG Evaluation using GPT-4 with Better Human Alignmentcited by 2,276
  3. 2023Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMscited by 974
  4. 2023Just Ask for Calibration: Strategies for Eliciting Calibrated Confidence Scores from Language Models Fine-Tuned with Human Feedbackcited by 812
  5. 2019Conformalized Quantile Regressioncited by 629
  6. 2021Learn then Test: Calibrating Predictive Algorithms to Achieve Risk Controlcited by 271
  7. 2023Conformal Language Modelingcited by 170
  8. 2025Phi-4-reasoning Technical Reportcited by 151
  9. 2023Conformal Prediction with Large Language Models for Multi-Choice Question Answeringcited by 139
  10. 2024Language Models with Conformal Factuality Guaranteescited by 131
  11. 2024VL-RewardBench: A Challenging Benchmark for Vision-Language Generative Reward Modelscited by 86
  12. 2021Conformal Prediction using Conditional Histogramscited by 77
21 more
  1. 2019A comparison of some conformal quantile regression methodscited by 76
  2. 2021Fair Conformal Predictors for Applications in Medical Imagingcited by 74
  3. 2024API Is Enough: Conformal Prediction for Large Language Models Without Logit-Accesscited by 69
  4. 2024Polos: Multimodal Metric Learning from Human Feedback for Image Captioningcited by 46
  5. 2023SocREval: Large Language Models with the Socratic Method for Reference-Free Reasoning Evaluationcited by 27
  6. 2024Boosted Conformal Prediction Intervalscited by 27
  7. 2025Analyzing Uncertainty of LLM-as-a-Judge: Interval Evaluations with Conformal Predictioncited by 27
  8. 2021Locally Valid and Discriminative Prediction Intervals for Deep Learning Modelscited by 24
  9. 2025Learning Conformal Abstention Policies for Adaptive Risk Management in Large Language and Vision-Language Modelscited by 21
  10. 2024Conformal Prediction via Regression-as-Classificationcited by 18
  11. 2024Black-box Uncertainty Quantification Method for LLM-as-a-Judgecited by 18
  12. 2026TRACER: Trajectory Risk Aggregation for Critical Episodes in Agentic Reasoningcited by 11
  13. 2025Learnable Conformal Prediction with Context-Aware Nonconformity Functions for Robotic Planning and Perceptioncited by 7
  14. 2024Navigating the Unknown: Uncertainty-Aware Compute-in-Memory Autonomy of Edge Roboticscited by 6
  15. 2025Beyond Confidence: Adaptive Abstention in Dual-Threshold Conformal Prediction for Autonomous System Perceptioncited by 6
  16. 2025Calibrated Decomposition of Aleatoric and Epistemic Uncertainty in Deep Features for Inference-Time Adaptationcited by 6
  17. 2025INTACT: Inducing Noise Tolerance through Adversarial Curriculum Training for LiDAR-based Safety-Critical Perception and Autonomycited by 5
  18. 2025Intelligent Sensing-to-Action for Robust Autonomy at the Edge: Opportunities and Challengescited by 5
  19. 2026Uncertainty-Guided Inference-Time Depth Adaptation for Transformer-Based Visual Trackingcited by 3
  20. 2026TRIAGE: Type-Routed Interventions via Aleatoric-Epistemic Gated Estimation in Robotic Manipulation and Adaptive Perception -- Don't Treat All Uncertainty the Samecited by 3
  21. 2026Belief Dynamics for Detecting Behavioral Shifts in Safe Collaborative Manipulationcited by 1

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 8.

8 papers.

  1. 2026PaintBench: Deterministic Evaluation of Precise Visual Editingcited by 2
  2. 2026Uncertainty Quantification for Computer-Use Agents: A Benchmark across Vision-Language Models and GUI Grounding Datasetscited by 2
  3. 2026MIBE: Multi-subject Interaction Benchmark and Evaluator for Personalized Image Generationcited by 2
  4. 2026CURA: Certified Runtime Alarms for Computer-Use Agentscited by 1
  5. 2026QVal: Cheaply Evaluating Dense Supervision Signals for Long-Horizon LLM Agentscited by 0
  6. 2026REMEDY: How Far Is Video Generation from Medical Education World Models?cited by 0
  7. 2026A Low Grounding Score Is Not an Ungrounded Judge: Identifying the Perceptibility Confound in Multimodal Oversightcited by 0
  8. 2026Beyond Plausibility: Verifiable Fine-Grained Image Editing on Structured Assetscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025The Art of Saying "Maybe": A Conformal Lens for Uncertainty Benchmarking in VLMs
  2. 2026Judge Model for Large-scale Multimodality Benchmarks
  3. 2026When Vision-Language Models Judge Without Seeing: Exposing Informativeness Bias
  4. 2026MJ1: Multimodal Judgment via Grounded Verification
  5. 2025Multi-Crit: Benchmarking Multimodal Judges on Pluralistic Criteria-Following
  6. 2026Bi-Level Prompt Optimization for Multimodal LLM-as-a-Judge
  7. 2024Multi-modal, Multi-task, Multi-criteria Automatic Evaluation with Vision Language Models
  8. 2026Visual Aesthetic Benchmark: Can Frontier Models Judge Beauty?