mlsanityData to 14 Oct 2026

Vision Language Models are Biased

arXiv 2505.23941 ↗First posted 29 May 2025cs.LG cs.CVcited by 66

On the map: Multimodal LLMs / vision-language models → VLM benchmarks and visual reasoning

Abstract (arXiv)

Large language models (LLMs) memorize a vast amount of prior knowledge from the Internet that helps them on downstream tasks but also may notoriously sway their outputs towards wrong or biased answers. In this work, we test how the knowledge about popular subjects hurt the accuracy of vision language models (VLMs) on standard, objective visual tasks of counting and identification. We find that state-of-the-art VLMs are strongly biased (e.g., unable to recognize the 4th stripe has been added to a 3-stripe Adidas logo) scoring an average of 17.05% accuracy in counting (e.g., counting stripes in an Adidas-like logo) across 7 diverse domains from animals, logos, chess, board games, optical illusions, to patterned grids. Removing image backgrounds nearly doubles accuracy (21.09 percentage points), revealing that contextual visual cues trigger these biased responses. Further analysis of VLMs' reasoning patterns shows that counting accuracy initially rises with thinking tokens, reaching ~40%, before declining with excessive reasoning. Our work presents an interesting failure mode in VLMs and a human-supervised automated framework for testing VLM biases. Code and data are available at: vlmsarebiased.github.io.

Key facts. Cited by 66 papers in the map (data through 14 Oct 2026). Placed in VLM benchmarks and visual reasoning, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 16 of its 93 references that are in the map. First posted on arXiv 29 May 2025.

Builds on

16 of 93 references.

Closest ancestors · 4

Same direction, newest first

  1. 2025Uncovering Cultural Representation Disparities in Vision-Language Models20 May
  2. 2024Probing Visual Language Priors in VLMs31 Dec
  3. 2024IllusionVQA: A Challenging Optical Illusion Dataset for Vision Language Models23 Mar
  4. 2023Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond24 Aug

Other references · 12

  1. 2025Qwen2.5-VL Technical Reportcited by 5,587
  2. 2024Qwen2 Technical Reportcited by 2,120
  3. 2024Chameleon: Mixed-Modal Early-Fusion Foundation Modelscited by 927
  4. 2024Pixtral 12Bcited by 179
  5. 2025MLLMs Know Where to Look: Training-free Perception of Small Visual Details with Multimodal LLMscited by 176
  6. 2024Forcing Diffuse Distributions out of Language Modelscited by 48
  7. 2024PhD: A ChatGPT-Prompted Visual hallucination Evaluation Datasetcited by 45
  8. 2024GenderBias-VL: Benchmarking Gender Bias in Vision Language Models via Counterfactual Probingcited by 31
  9. 2023Survey of Social Bias in Vision-Language Modelscited by 26
  10. 2024Heads, Tails, and AI Fails: LLMs, Randomness, and Human Judgmentscited by 26
  11. 2024Investigating Implicit Bias in Large Language Models: A Large-Scale Study of Over 50 LLMscited by 25
  12. 2023GPTBIAS: A Comprehensive Framework for Evaluating Bias in Large Language Modelscited by 16

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 17; 2026: 49.

66 papers.

Sort:

By area:

  1. 2026MIRAGE: The Illusion of Visual Understandingcited by 65
  2. 2026Seed1.8 Model Card: Towards Generalized Real-World Agencycited by 50
  3. 2025Understanding Language Prior of LVLMs by Contrasting Chain-of-Embeddingcited by 14
  4. 2025Your other Left! Vision-Language Models Fail to Identify Relative Positions in Medical Imagescited by 13
  5. 2026VLMs Need Words: Vision Language Models Ignore Visual Detail In Favor of Semantic Anchorscited by 13
  6. 2025EdiVal-Agent: An Object-Centric Framework for Automated, Fine-Grained Evaluation of Multi-Turn Editingcited by 10
  7. 2026Do 3D Large Language Models Really Understand 3D Spatial Relationships?cited by 9
  8. 2026Video Models Reason Early: Exploiting Plan Commitment for Maze Solvingcited by 9
  9. 2026Do VLMs Perceive or Recall? Probing Visual Perception vs. Memory with Classic Visual Illusionscited by 8
  10. 2026SPARC: Separating Perception And Reasoning Circuits for Test-time Scaling of VLMscited by 7
  11. 2025Same or Not? Enhancing Visual Perception in Vision-Language Modelscited by 5
  12. 2026Mechanisms of Prompt-Induced Hallucination in Vision-Language Modelscited by 5
38 more
  1. 2026Watch Before You Answer: Learning from Visually Grounded Post-Trainingcited by 5
  2. 2025VisualOverload: Probing Visual Understanding of VLMs in Really Dense Scenescited by 4
  3. 2026VISTAQA: Benchmarking Joint Visual Question Answering and Pixel-Level Evidencecited by 4
  4. 2026PerceptionBench: Evaluating Atomic Visual Perception in Multimodal Large Language Modelscited by 4
  5. 2025Vision Language Models are Confused Touristscited by 3
  6. 20256 Fingers, 1 Kidney: Natural Adversarial Medical Images Reveal Critical Weaknesses of Vision-Language Modelscited by 3
  7. 2025VisRes Bench: On Evaluating the Visual Reasoning Capabilities of VLMscited by 3
  8. 2026Mitigating Perceptual Judgment Bias in Multimodal LLM-as-a-Judge via Perceptual Perturbation and Reward Modelingcited by 3
  9. 2025VLM-SlideEval: Evaluating VLMs on Structured Comprehension and Perturbation Sensitivity in PPTcited by 2
  10. 2026Do VLMs Read or Rewrite? On Transcription Faithfulness in Vision-Language Modelscited by 2
  11. 2025CAD2DMD-SET: Synthetic Generation Tool of Digital Measurement Device CAD Model Datasets for fine-tuning Large Vision-Language Modelscited by 1
  12. 2025Surgeons Are Indian Males and Speech Therapists Are White Females: Auditing Biases in Vision-Language Models for Healthcare Professionalscited by 1
  13. 2025MMRPT: MultiModal Reinforcement Pre-Training via Masked Vision-Dependent Reasoningcited by 1
  14. 2026ArtifactLens: Hundreds of Labels Are Enough for Artifact Detection with VLMscited by 1
  15. 2026Stress Tests REVEAL Fragile Temporal and Visual Grounding in Video-Language Modelscited by 1
  16. 2026A Closed-Form Solution for Debiasing Vision-Language Models with Utility Guarantees Across Modalities and Taskscited by 1
  17. 2026Beyond Perception Errors: Semantic Fixation in Large Vision-Language Modelscited by 1
  18. 2026Do Physics Foundation Models Learn Generalizable Physics? A Bias-Aware Benchmark Across Physical Regimes and Distribution Shiftscited by 1
  19. 2026Vision-Language Models Suppress Female Representations Under Ambiguous Inputcited by 1
  20. 2026MT-EditFlow: Reinforcement Learning for Multi-Turn Image Editing with Flow Matchingcited by 1
  21. 2026The Count Is There, but Misaligned: Understanding and Correcting Counting Failures in VLMscited by 1
  22. 2026HoloGeo: Mitigating Landmark Bias in Geo-localization via Evidence-Driven Reasoningcited by 1
  23. 2026Image-Space Rule Discoverycited by 1
  24. 2026Hard Vision, Easy Vision: What GPT-6 Astra Reveals Across Computer Visioncited by 1
  25. 2025VMMU: A Vietnamese Multitask Multimodal Understanding and Reasoning Benchmarkcited by 0
  26. 2025Modal Aphasia: Can Unified Multimodal Models Describe Images From Memory?cited by 0
  27. 2025Context informs pragmatic interpretation in vision-language modelscited by 0
  28. 2025INQUIRE-Search: Interactive Discovery in Large-Scale Biodiversity Databasescited by 0
  29. 2025Probing the effectiveness of World Models for Spatial Reasoning through Test-time Scalingcited by 0
  30. 2026VLM-CAD: VLM-Optimized Collaborative Agent Design Workflow for Analog Circuit Sizingcited by 0
  31. 2026KidVis: Do Multimodal Large Language Models Possess the Visual Perceptual Capabilities of a 6-Year-Old?cited by 0
  32. 2026Spatial Colour Mixing Illusions as a Perception Stress Test for Vision-Language Modelscited by 0
  33. 2026MuRF: Unlocking the Multi-Scale Potential of Vision Foundation Modelscited by 0
  34. 2026Semantic Richness or Geometric Reasoning? The Fragility of VLM's Visual Invariancecited by 0
  35. 2026NearID: Identity Representation Learning via Near-identity Distractorscited by 0
  36. 2026Bridging Foundation Models and ASTM Metallurgical Standards for Automated Grain Size Estimation from Microscopy Imagescited by 0
  37. 2026FIND: Toward Multimodal Financial Reasoning and Question Answering for Indic Languagescited by 0
  38. 2026VLMs Trace Without Tracking: Diagnosing Failures in Visual Path Followingcited by 0

50 most cited of 66 All 66 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2025Can Vision-Language Models Count? A Synthetic Benchmark and Analysis of Attention-Based Interventions
  2. 2025Mixed Signals: Decoding VLMs' Reasoning and Underlying Bias in Vision-Language Conflict
  3. 2026CounterCount: A Diagnostic Framework for Counting Bias in Vision Language Models
  4. 2026Scale Can't Overcome Pragmatics: The Impact of Reporting Bias on Vision-Language Reasoning
  5. 2026Counting Circuits: Mechanistic Interpretability of Visual Reasoning in Large Vision-Language Models
  6. 2025[De|Re]constructing VLMs' Reasoning in Counting
  7. 2026Counting to Four is still a Chore for VLMs
  8. 2025Examining Vision Language Models through Multi-dimensional Experiments with Vision and Text Features