# Visually Prompted Benchmarks Are Surprisingly Fragile

arXiv: https://arxiv.org/abs/2512.17875 · HTML: https://mlsanity.com/paper/2512.17875/

## Key facts

- Cited by 7 papers in the map (data through 14 Oct 2026).
- Builds on 27 of its 32 references that are in the map.
- First posted on arXiv 19 Dec 2025.

## Builds on (most cited first, 27 of 27)

- [The Unreasonable Effectiveness of Deep Features as a Perceptual Metric](https://mlsanity.com/paper/1801.03924/) (2018)
- [Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotations](https://mlsanity.com/paper/1602.07332/) (2016)
- [MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGI](https://mlsanity.com/paper/2311.16502/) (2023)
- [MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models](https://mlsanity.com/paper/2306.13394/) (2023)
- [CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning](https://mlsanity.com/paper/1612.06890/) (2016)
- [Depth Anything V2](https://mlsanity.com/paper/2406.09414/) (2024)
- [Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting](https://mlsanity.com/paper/2310.11324/) (2023)
- [SpatialVLM: Endowing Vision-Language Models with Spatial Reasoning Capabilities](https://mlsanity.com/paper/2401.12168/) (2024)
- [BLINK: Multimodal Large Language Models Can See but Not Perceive](https://mlsanity.com/paper/2404.12390/) (2024)
- [From Crowdsourced Data to High-Quality Benchmarks: Arena-Hard and BenchBuilder Pipeline](https://mlsanity.com/paper/2406.11939/) (2024)
- [A Corpus for Reasoning About Natural Language Grounded in Photographs](https://mlsanity.com/paper/1811.00491/) (2018)
- [Visual7W: Grounded Question Answering in Images](https://mlsanity.com/paper/1511.03416/) (2015)
- [SpatialRGPT: Grounded Spatial Reasoning in Vision Language Models](https://mlsanity.com/paper/2406.01584/) (2024)
- [What's "up" with vision-language models? Investigating their struggle with spatial reasoning](https://mlsanity.com/paper/2310.19785/) (2023)
- [What does CLIP know about a red circle? Visual prompt engineering for VLMs](https://mlsanity.com/paper/2304.06712/) (2023)
- [SPair-71k: A Large-scale Benchmark for Semantic Correspondence](https://mlsanity.com/paper/1908.10543/) (2019)
- [DynaMath: A Dynamic Visual Benchmark for Evaluating Mathematical Reasoning Robustness of Vision Language Models](https://mlsanity.com/paper/2411.00836/) (2024)
- [When Benchmarks are Targets: Revealing the Sensitivity of Large Language Model Leaderboards](https://mlsanity.com/paper/2402.01781/) (2024)
- [Is A Picture Worth A Thousand Words? Delving Into Spatial Reasoning for Vision Language Models](https://mlsanity.com/paper/2406.14852/) (2024)
- [Perception Tokens Enhance Visual Reasoning in Multimodal Language Models](https://mlsanity.com/paper/2412.03548/) (2024)
- [The Leaderboard Illusion](https://mlsanity.com/paper/2504.20879/) (2025)
- [Quantifying Variance in Evaluation Benchmarks](https://mlsanity.com/paper/2406.10229/) (2024)
- [InternSpatial: A Comprehensive Dataset for Spatial Reasoning in Vision-Language Models](https://mlsanity.com/paper/2506.18385/) (2025)
- [MVTamperBench: Evaluating Robustness of Vision-Language Models](https://mlsanity.com/paper/2412.19794/) (2024)
- [RoCOCO: Robustness Benchmark of MS-COCO to Stress-test Image-Text Matching Models](https://mlsanity.com/paper/2304.10727/) (2023)
- [Unveiling the Lack of LVLM Robustness to Fundamental Visual Variations: Why and Path Forward](https://mlsanity.com/paper/2504.16727/) (2025)
- [On the Reliability of Vision-Language Models Under Adversarial Frequency-Domain Perturbations](https://mlsanity.com/paper/2507.22398/) (2025)

## Built on by (most cited first, 7 of 7)

- [When the Prompt Becomes Visual: Vision-Centric Jailbreak Attacks for Large Image Editing Models](https://mlsanity.com/paper/2602.10179/) (2026)
- [VPA-Guard: Defending and Benchmarking Image-to-Video Generation Against Visual Prompt Attacks](https://mlsanity.com/paper/2606.25592/) (2026)
- [Semantic Richness or Geometric Reasoning? The Fragility of VLM's Visual Invariance](https://mlsanity.com/paper/2604.01848/) (2026)
- [The Cost of Language: Centroid Erasure Exposes and Exploits Modal Competition in Multimodal Language Models](https://mlsanity.com/paper/2604.14363/) (2026)
- [Compositionality Emerges in a Narrow Depth-Connectivity Regime: Architecture Constraints and Solution Manifolds](https://mlsanity.com/paper/2606.19941/) (2026)
- [Multimodal Thinking with Renderable Programs](https://mlsanity.com/paper/2609.30130/) (2026)
- [SLVR: Structured Latent Visual Reasoning via Human-like Reasoning Flows](https://mlsanity.com/paper/2610.10563/) (2026)

## Abstract (arXiv)

A key challenge in evaluating VLMs is testing models' ability to analyze visual content independently from their textual priors. Recent benchmarks such as BLINK probe visual perception through visual prompting, where questions about visual content are paired with coordinates to which the question refers, with the coordinates explicitly marked in the image itself. While these benchmarks are an important part of VLM evaluation, we find that existing models are surprisingly fragile to seemingly irrelevant details of visual prompting: simply changing a visual marker from red to blue can completely change rankings among models on a leaderboard. By evaluating nine commonly-used open- and closed-source VLMs on two visually prompted tasks, we demonstrate how details in benchmark setup, including visual marker design and dataset size, have a significant influence on model performance and leaderboard rankings. These effects can even be exploited to lift weaker models above stronger ones; for instance, slightly increasing the size of the visual marker results in open-source InternVL3-8B ranking alongside or better than much larger proprietary models like Gemini 2.5 Pro. We further show that low-level inference choices that are often ignored in benchmarking, such as JPEG compression levels in API calls, can also cause model lineup changes. These details have substantially larger impacts on visually prompted benchmarks than on conventional semantic VLM evaluations. To mitigate this instability, we curate existing datasets to create VPBench, a larger visually prompted benchmark with 16 visual marker variants. We open-source VPBench and our analysis framework at: https://lisadunlap.github.io/vpbench/.