# Auditing VLM Benchmarks from Existing Evaluation Runs

arXiv: https://arxiv.org/abs/2610.13548 · HTML: https://mlsanity.com/paper/2610.13548/

## 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 20 of its 21 references that are in the map.
- First posted on arXiv 9 Oct 2026.

## Placed in

- Multimodal LLMs / vision-language models / [VLM benchmarks and visual reasoning](https://mlsanity.com/mllm/vlm-benchmarks-and-visual-reasoning/)

## Builds on (most cited first, 20 of 20)

- [Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering](https://mlsanity.com/paper/1612.00837/) (2016)
- [MMBench: Is Your Multi-modal Model an All-around Player?](https://mlsanity.com/paper/2307.06281/) (2023)
- [Are We on the Right Way for Evaluating Large Vision-Language Models?](https://mlsanity.com/paper/2403.20330/) (2024)
- [HallusionBench: An Advanced Diagnostic Suite for Entangled Language Hallucination and Visual Illusion in Large Vision-Language Models](https://mlsanity.com/paper/2310.14566/) (2023)
- [Large Language Models Are Not Robust Multiple Choice Selectors](https://mlsanity.com/paper/2309.03882/) (2023)
- [VLMEvalKit: An Open-Source Toolkit for Evaluating Large Multi-Modality Models](https://mlsanity.com/paper/2407.11691/) (2024)
- [Pervasive Label Errors in Test Sets Destabilize Machine Learning Benchmarks](https://mlsanity.com/paper/2103.14749/) (2021)
- [LMMs-Eval: Reality Check on the Evaluation of Large Multimodal Models](https://mlsanity.com/paper/2407.12772/) (2024)
- [Are We Done with MMLU?](https://mlsanity.com/paper/2406.04127/) (2024)
- [NaturalBench: Evaluating Vision-Language Models on Natural Adversarial Samples](https://mlsanity.com/paper/2410.14669/) (2024)
- [Do Large Language Model Benchmarks Test Reliability?](https://mlsanity.com/paper/2502.03461/) (2025)
- [Correlated Errors in Large Language Models](https://mlsanity.com/paper/2506.07962/) (2025)
- [Redundancy Principles for MLLMs Benchmarks](https://mlsanity.com/paper/2501.13953/) (2025)
- [MMEvalPro: Calibrating Multimodal Benchmarks Towards Trustworthy and Efficient Evaluation](https://mlsanity.com/paper/2407.00468/) (2024)
- [Fantastic Bugs and Where to Find Them in AI Benchmarks](https://mlsanity.com/paper/2511.16842/) (2025)
- [Both Text and Images Leaked! A Systematic Analysis of Data Contamination in Multimodal LLM](https://mlsanity.com/paper/2411.03823/) (2024)
- [DatBench: Discriminative, Faithful, and Efficient VLM Evaluations](https://mlsanity.com/paper/2601.02316/) (2026)
- [Information Density Principle for MLLM Benchmarks](https://mlsanity.com/paper/2503.10079/) (2025)
- [Do Vision-Language Models See or Guess? Measuring and Reducing Textual-Prior Reliance with a Phrasing-Controlled Benchmark](https://mlsanity.com/paper/2606.10400/) (2026)
- [BenchMarker: An Education-Inspired Toolkit for Highlighting Flaws in Multiple-Choice Benchmarks](https://mlsanity.com/paper/2602.06221/) (2026)

## Abstract (arXiv)

Every evaluation run of a vision-language model (VLM) already produces per-question artifacts (predictions, judged correctness, metadata) that leaderboards reduce to a single score and then discard. We reuse those artifacts to audit the benchmark itself, and add an audit layer to VLMEvalKit: six detectors that run as post-processing over saved outputs, plus one additional text-only rerun with visual inputs removed, emitting a per-question report. The layer reports per-question findings together with aggregate audit statistics; for the headline quantities, it attaches permutation-null excess where applicable, interval estimates for proportions, and sensitivity to the model pool. Auditing MMStar, a benchmark curated specifically so that the image is indispensable, with three models, we find zero aggregate accuracy gain from the image on 20.2% of its questions. An item-matched permutation null accounts for 17.0%, leaving a +3.2 pp excess (corrected permutation p = 0.001). This excess is concentrated in the mathematics and science categories and is absent from coarse perception, a localization consistent with knowledge substituting for visual evidence; confirming that mechanism would require item-level review. A second detector flags 134 questions on which every model that answered in parseable form picked the same non-gold option, and these candidates peak in the perception categories, whereas the zero-gain excess is largest in mathematics and science & technology. Replacing the LLM judge with exact string matching changes no conclusion, and every scoring discrepancy between the two traces to one model whose free-form answers a parser cannot read. On a binary yes/no benchmark the detectors measurably fall out of calibration, a scope boundary that we measure and report. The audit layer, the prediction artifacts and the generated reports accompany this submission.