mlsanityData to 14 Oct 2026

The Last Visible Pixel: Probing Fine-Scale Perception in Vision-Language Models

arXiv 2606.07861 ↗First posted 5 Jun 2026cs.CV cs.AIcited by 0

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

Abstract (arXiv)

Recent vision-language models (VLMs) excel at multimodal understanding and reasoning, yet their fine-grained visual perception remains underexplored. A natural extension of ``How many r are there in Strawberry?'' asks: how small a visual pattern can a VLM reliably perceive? As such, we introduce FineSightBench, a new benchmark that systematically probes this limit by separating perception tasks (pixel-level recognition of letters, shapes, objects) from reasoning tasks (spatial reasoning, counting, ordering over small targets) across controlled scales of 4--48px. Through comprehensive experiments and detailed failure mode analysis on state-of-the-art models, we reveal a sharp dissociation: perception saturates around 12px, while reasoning remains limited even at larger scales, with persistent numeracy and sequence errors. These findings expose fundamental deficiencies in VLMs' fine-scale visual reasoning that demand more rigorous evaluation.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in VLM benchmarks and visual reasoning, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 25 of its 35 references that are in the map. First posted on arXiv 5 Jun 2026.

Builds on

25 of 35 references.

Closest ancestors · 6

Same direction, newest first

  1. 2025VLMs have Tunnel Vision: Evaluating Nonlocal Visual Reasoning in Leading VLMs4 Jul
  2. 2025Native Visual Understanding: Resolving Resolution Dilemmas in Vision-Language Models15 Jun
  3. 2025Unveiling the Lack of LVLM Robustness to Fundamental Visual Variations: Why and Path Forward23 Apr
  4. 2024Vision language models are blind: Failing to translate detailed visual features into words9 Jul
  5. 2023MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contexts3 Oct
  6. 2023Kosmos-2: Grounding Multimodal Large Language Models to the World26 Jun

Other references · 19

  1. 2020Language Models are Few-Shot Learnerscited by 32,419
  2. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 30,186
  3. 2024Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolutioncited by 4,234
  4. 2023Sigmoid Loss for Language Image Pre-Trainingcited by 3,109
  5. 2023MMBench: Is Your Multi-modal Model an All-around Player?cited by 2,141
  6. 2023Evaluating Object Hallucination in Large Vision-Language Modelscited by 1,729
  7. 2024Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scalingcited by 1,599
  8. 2023SEED-Bench: Benchmarking Multimodal LLMs with Generative Comprehensioncited by 940
  9. 2023Language Is Not All You Need: Aligning Perception with Language Modelscited by 543
  10. 2016Synthetic Data for Text Localisation in Natural Imagescited by 396
  11. 2024SpatialRGPT: Grounded Spatial Reasoning in Vision Language Modelscited by 311
  12. 2023Patch n' Pack: NaViT, a Vision Transformer for any Aspect Ratio and Resolutioncited by 229
7 more
  1. 2025MLLMs Know Where to Look: Training-free Perception of Small Visual Details with Multimodal LLMscited by 176
  2. 2024SEED-Bench-2-Plus: Benchmarking Multimodal Large Language Models with Text-Rich Visual Comprehensioncited by 146
  3. 2024Is A Picture Worth A Thousand Words? Delving Into Spatial Reasoning for Vision Language Modelscited by 113
  4. 2022Language Models are General-Purpose Interfacescited by 97
  5. 2023SEED-Bench-2: Benchmarking Multimodal Large Language Modelscited by 97
  6. 2025Exploring the Impact of Temperature on Large Language Models:Hot or Cold?cited by 40
  7. 2025WearVQA: A Visual Question Answering Benchmark for Wearables in Egocentric Authentic Real-world scenarioscited by 9

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025VisRes Bench: On Evaluating the Visual Reasoning Capabilities of VLMs
  2. 2023Synthesize, Diagnose, and Optimize: Towards Fine-Grained Vision-Language Understanding
  3. 2026Do Vision-Language Models Truly Perform Vision Reasoning? A Rigorous Study of the Modality Gap
  4. 2024UniBench: Visual Reasoning Requires Rethinking Vision-Language Beyond Scaling
  5. 2026Don't Show Pixels, Show Cues: Unlocking Visual Tool Reasoning in Language Models via Perception Programs
  6. 2025HueManity: Probing Fine-Grained Visual Perception in MLLMs
  7. 2026PerceptionBench: Evaluating Atomic Visual Perception in Multimodal Large Language Models
  8. 2026From Perception to Integration: Revisiting the Internal Dynamics of Reasoning in Vision-Language Models