mlsanityData to 7 Oct 2026

VLM benchmarks and visual reasoning

2,063 papers5 this weekLow confidence in this grouping: the papers may belong to more than one topic, or the group may changeLow confidence in this grouping: the papers may belong to more than one topic, or the group may change · About

Keywords: visual, reasoning, vlms, vision-language, language

Key facts. VLM benchmarks and visual reasoning: 2,063 papers on arXiv since 2021 (data through 7 Oct 2026). 265 of its 2,063 mapped papers (13%) cite “Flamingo: a Visual Language Model for Few-Shot Learning” (2022, arXiv 2204.14198). 256 of its 2,063 mapped papers (12%) cite “MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contexts” (2023, arXiv 2310.02255). 65 new papers 17 Sep–7 Oct.

What’s new

17 Sep – 7 Oct 2026

65 new papers, 17 Sep – 7 Oct; 5 this week.

  1. This week · 5
  2. Visual Evidence Under Cross-Examination: Evaluating and Controlling Decision-Level Evidence Use in Vision-Language Models
  3. GeoPID: Decomposing and Steering Visual Information in Vision-Language Models also fits: Embodied and 3D vision-language models
  4. Matching Object or Relation? Tracing Abstract Reasoning Inside VLMs also fits: Embodied and 3D vision-language models
  5. Harnessing Multimodal Large Language Models for Training-Free Human-Object Interaction Detection also fits: Embodied and 3D vision-language models
  6. Readout Blindness: VLM Scores Miss the Spatial Direction Their Frozen Encoders Retain also fits: Prompt learning and adaptation of CLIP
  7. Earlier · 60
  8. When to Rethink: Learning Multi-Perspective Self-Verification for Vision-Language Models also fits: MLLM training and visual instruction tuning
  9. Atomic Visual Entailment: Enhancing Zero-Shot Vision-Language Reasoning through Atomic Fact Decomposition and Learned Selection
  10. ReMAP: Restoring the Perceptual Cycle with Reasoning-Time Latent Visual Memory
  11. Look Where You Say You're Looking: Self-Grounded Attention for Visual Reasoning
  12. Trinity: Self-Evolving Vision-Language Models with a Self-Verifier also fits: Embodied and 3D vision-language models
All 65
  1. KVE-KD: Key Visual Evidence-Guided Knowledge Distillation for Vision-Language Models
  2. Evaluating VQA in Vision Language Models using Cooperative Principles
  3. SLVR: Structured Latent Visual Reasoning via Human-like Reasoning Flows
  4. Not All Error Yields to Scale: Where Scaling Stops in Vision-Language Inference also fits: Visual token pruning and efficiency
  5. When the Judge Acts: Auditing VLM-Guided Image Selection on Culturally Situated Prompts also fits: Social bias and fairness in VLMs
  6. Skeleton-and-Strategy Prompting: Training-Free Negation Understanding for Vision-Language Models
  7. Geometric Similarity in VLM Low-Level Vision Representations also fits: Prompt learning and adaptation of CLIP
  8. VisionQ: VLM-as-a-Judge Taxonomy, Dataset and Benchmark for Qualitative Analysis in Computer Vision
  9. Aligning Thoughts with Answers: Probability Rewards to Tame Thinking Drift also fits: Embodied and 3D vision-language models
  10. Reinforcing Multimodal Reasoning via Token-Level Perception-Grounded Advantage Estimation also fits: MLLM training and visual instruction tuning
  11. Distill the Visual Evidence, Not Just the Answer: Cross-World On-Policy Distillation for Vision-Language Models
  12. Revisit to Segment: Working Memory Distillation for Reasoning Segmentation also fits: Embodied and 3D vision-language models
  13. Composition, Not Conversation: VLMs Lose the Scene, Not the Thread
  14. Inductive Visual Logic for Few-Shot Out-Of-Distribution Adaptation in VLMs also fits: Prompt learning and adaptation of CLIP
  15. From Routing Signals to Selective Review: Visual regrounding in MoE VLMs
  16. NeuronEye: Query-Guided Visual Concept Activation for Vision-Language Reasoning also fits: Prompt learning and adaptation of CLIP
  17. It's Not What the Image Shows: Irrelevant Context Destabilises VLM Judges Without Informing Them also fits: Social bias and fairness in VLMs
  18. CHOQOLATE: Organizing Concept Bottleneck Latent Spaces with Choquet Integrals also fits: Chart understanding and anomaly detection
  19. Tracing the Evidence: Faithful Token Attribution Through Vision-Language Reasoning
  20. EviViT: Evidence-Adaptive Vision Transformers for Fine-Grained Perception also fits: Visual token pruning and efficiency
  21. Benchmarking Vision-Language Models on Synapse Detection and Proofreading in Connectomics also fits: Prompt learning and adaptation of CLIP
  22. Question-Specific Knowledge Graphs for Efficient Visual Reasoning
  23. From Perception to Integration: Revisiting the Internal Dynamics of Reasoning in Vision-Language Models
  24. When VLMs Trust Context: Evaluating Scene Text Recognition under Misleading Context also fits: Document understanding and multimodal RAG
  25. Rethinking Latent Visual Reasoning: Grounding Latent Reasoning in Visual Evidence
  26. Distilling Visual Reasoning into Text Space
  27. SCOPD: Sparse-Context On-Policy Self-Distillation for Efficient Vision-Language Models also fits: Visual token pruning and efficiency
  28. Program-Verified Self-Evolution for Vision-Language Models
  29. Seeing and Solving Are Not Enough for Vision-Language Models
  30. MoGround: Measuring and Mitigating Modality Distraction in Vision-Language Models also fits: MLLM training and visual instruction tuning
  31. TTRSD: Test-Time Reinforcement Learning with Self-Distillation for Vision-Language Models also fits: Embodied and 3D vision-language models
  32. OmniMoE-VL: A Sparse Vision-Language Model with Coupled Visual-Depth Routing also fits: Visual token pruning and efficiency
  33. MM-OPD: Towards One More Bottleneck Between Perception and Reasoning
  34. From Knowing to Abstaining: Bridging the Representation-Action Gap in Vision-Language Models
  35. When Helpful Text Hurts: Option-Redirecting Bias in Vision-Language Models
  36. Progressive-View On-Policy Distillation for Regional-to-Global Transfer in Multimodal LLMs also fits: MLLM training and visual instruction tuning
  37. SynDORBench: Evaluating LVLM Perceptual Robustness Under Physically Constrained Visibility Conditions also fits: Embodied and 3D vision-language models
  38. Diagnosing the Sources of Compositional Failure in Vision-Language Models: A Controlled Analysis
  39. Can Linguistic Reasoning Vectors Enhance Multimodal Reasoning Ability? also fits: MLLM training and visual instruction tuning
  40. FLIP: Final Layer Inference-Time Probing for Vision-Language Models also fits: Embodied and 3D vision-language models
  41. CCRV-Bench: Constraint-Based Evaluation of Causal Reasoning in Vision-Language Models
  42. Skip the Talk, Re-Focus on Vision: Latent Reasoning for Reasoning Segmentation in Multimodal Large Language Models
  43. Multimodal Thinking with Renderable Programs
  44. Mind What Matters for Reasoning: Aligning Cross-Modal Attention via Selective Probability Mass Concentration also fits: MLLM training and visual instruction tuning
  45. Looks the Same, Answers Differently: Flip-Direction Steering for Robust Vision-Language Reasoning also fits: Embodied and 3D vision-language models
  46. What Looks Like a Capability Limit in Vision-Language Models Is a Readout Limit also fits: Prompt learning and adaptation of CLIP
  47. PRISM-VLM: A Multi-Axis Discriminative Benchmark for Compact Vision-Language Models also fits: Prompt learning and adaptation of CLIP
  48. Reading Right, Answering Wrong: How Visual Configuration Changes Affect Evidence Use in VLMs
  49. Look Where It Counts: A Free, Label-Free Visual Evidence Signal for Fine-Grained Vision-Language Reasoning
  50. MinCU: A Fine-Grained Benchmark for Grounded Minimal-Change Understanding in Image Pairs also fits: Image editing and quality assessment
  51. CrowdCue: Specialist-Cue Conditioning for Vision-Language Crowd Counting also fits: Visual token pruning and efficiency
  52. Seeing is not Enough: Vision-Language Models Perceive Evidence but Fail to Act
  53. DiaVLo: Diagnosing Behaviours of Vision-Language Models also fits: Embodied and 3D vision-language models
  54. Benchmarking the Explanatory Quality of Open-Weight Vision-Language Models in Face Recognition also fits: Social bias and fairness in VLMs
  55. Cross-Modal Attention Acts as a Frequency Filter: Why Verbose Prompts Improve Robustness in Vision-Language Models

Foundations

Most cited by papers here

  1. Flamingo: a Visual Language Model for Few-Shot Learning2022cited by 265 here
  2. MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contexts2023cited by 256 here
  3. Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond2023cited by 227 here
  4. BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation2022cited by 199 here
  5. Are We on the Right Way for Evaluating Large Vision-Language Models?2024cited by 146 here

Lineage

The papers this direction builds on, oldest first.

Each paper on the line cites the one before.

  1. 2022BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generationcited by 199 here
  2. 2022Flamingo: a Visual Language Model for Few-Shot Learningcited by 265 here
  3. 2023InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuningcited by 247 here
  4. 2023Evaluating Object Hallucination in Large Vision-Language Modelscited by 171 here
  5. 2023MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Modelscited by 145 here
  6. 2023SEED-Bench: Benchmarking Multimodal LLMs with Generative Comprehensioncited by 108 here
  7. 2023Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyondcited by 227 here
  8. 2024Are We on the Right Way for Evaluating Large Vision-Language Models?cited by 146 here
  9. 2024Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolutioncited by 249 here
  10. 2024Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scalingcited by 111 here
14 more it builds on, not in this chain
  1. 2023Kosmos-2: Grounding Multimodal Large Language Models to the Worldcited by 96 here
  2. 2023MMBench: Is Your Multi-modal Model an All-around Player?cited by 230 here
  3. 2023MM-Vet: Evaluating Large Multimodal Models for Integrated Capabilitiescited by 144 here
  4. 2023MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contextscited by 256 here
  5. 2023Improved Baselines with Visual Instruction Tuningcited by 356 here
  6. 2023HallusionBench: An Advanced Diagnostic Suite for Entangled Language Hallucination and Visual Illusion in Large Vision-Language Modelscited by 107 here
  7. 2023V*: Guided Visual Search as a Core Mechanism in Multimodal LLMscited by 116 here
  8. 2023InternVL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic Taskscited by 190 here
  9. 2024Eyes Wide Shut? Exploring the Visual Shortcomings of Multimodal LLMscited by 109 here
  10. 2024Measuring Multimodal Mathematical Reasoning with MATH-Vision Datasetcited by 120 here
  11. 2024MathVerse: Does Your Multi-modal LLM Truly See the Diagrams in Visual Math Problems?cited by 114 here
  12. 2024BLINK: Multimodal Large Language Models Can See but Not Perceivecited by 102 here
  13. 2024LLaVA-OneVision: Easy Visual Task Transfercited by 189 here
  14. 2025Qwen3-VL Technical Reportcited by 120 here

The 24 most cited of 2,835.

Browse

30 of 2,063, ranked 6–35

  1. 6Measuring Multimodal Mathematical Reasoning with MATH-Vision Dataset2024120
  2. 7MathVerse: Does Your Multi-modal LLM Truly See the Diagrams in Visual Math Problems?2024114
  3. 8Kosmos-2: Grounding Multimodal Large Language Models to the World202396
  4. 9Visual CoT: Advancing Multi-Modal Language Models with a Comprehensive Dataset and Benchmark for Chain-of-Thought Reasoning202493
  5. 10LLaVA-CoT: Let Vision Language Models Reason Step-by-Step202486
  6. 11Visual Sketchpad: Sketching as a Visual Chain of Thought for Multimodal Language Models202472
  7. 12SimVLM: Simple Visual Language Model Pretraining with Weak Supervision202168
  8. 13LISA: Reasoning Segmentation via Large Language Model202366
  9. 14Vision-Language Models for Vision Tasks: A Survey202356
  10. 15Ferret: Refer and Ground Anything Anywhere at Any Granularity202352
All 30
  1. 16Compositional Chain-of-Thought Prompting for Large Multimodal Models202351
  2. 17What matters when building vision-language models?202449
  3. 18Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Vision-Language Models202449
  4. 19GLaMM: Pixel Grounding Large Multimodal Model202349
  5. 20OpenFlamingo: An Open-Source Framework for Training Large Autoregressive Vision-Language Models202347
  6. 21Machine Mental Imagery: Empower Multimodal Reasoning with Latent Visual Tokens202546
  7. 22We-Math: Does Your Large Multimodal Model Achieve Human-like Mathematical Reasoning?202441
  8. 23Imagine while Reasoning in Space: Multimodal Visualization-of-Thought202538
  9. 24PaliGemma: A versatile 3B VLM for transfer202437
  10. 25Kimi-VL Technical Report202536
  11. 26Math-LLaVA: Bootstrapping Mathematical Reasoning for Multimodal Large Language Models202435
  12. 27VLMo: Unified Vision-Language Pre-Training with Mixture-of-Modality-Experts202135
  13. 28Multi-Grained Vision Language Pre-Training: Aligning Texts with Visual Concepts202135
  14. 29Vision language models are blind: Failing to translate detailed visual features into words202434
  15. 30Multimodal Chain-of-Thought Reasoning: A Comprehensive Survey202533
  16. 31Latent Visual Reasoning202533
  17. 32Seeing Out of tHe bOx: End-to-End Pre-training for Vision-Language Representation Learning202131
  18. 33FILIP: Fine-grained Interactive Language-Image Pre-Training202131
  19. 34ReFocus: Visual Editing as a Chain of Thought for Structured Image Understanding202529
  20. 35VisuLogic: A Benchmark for Evaluating Visual Reasoning in Multi-modal Large Language Models202528