mlsanityData to 7 Oct 2026

Unexplored flaws in multiple-choice VQA make benchmarking unreliable

arXiv 2511.22341 ↗First posted 27 Nov 2025cs.CV cs.LGcited by 1

Not on the map. Why?

Abstract (arXiv)

Previous works identify sensitivity to option order as a key issue in multiple-choice VQA (MC-VQA) evaluation and propose protocols to mitigate this effect. We show that such mitigation is insufficient to ensure the validity of MC-VQA as a reliable benchmark for Multimodal Large Language Model (MLLMs): performance remains highly sensitive to semantically neutral prompt format choices that are not controlled by current benchmarks. In a large-scale study spanning seven MLLMs and five MC-VQAs datasets, we find frequent rank reversals even under order-invariant evaluation. These reversals arise when we systematically vary option ID sets, delimiters, and separators, yielding 48 semantically equivalent prompt formats. Mechanistic analyses trace this instability to low-level language modeling effects: tokenizer-induced fusion or removal of option ID tokens introduces corrupted option ID tokens into the input sequence, while the choice of option ID sets directly affects the reliability of attention patterns for option selection. Accordingly, MC-VQA rankings correlate weakly with open-ended evaluation, indicating that MC-VQA reflects option-selection dynamics in addition to multimodal reasoning. These findings identify prompt formatting as a major, previously under-controlled confounder in MC-VQA benchmarking and motivate evaluation protocols that explicitly control prompt format sensitivity.

Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Builds on 40 of its 53 references that are in the map. First posted on arXiv 27 Nov 2025.

Builds on

40 of 53 references.

References · 40

  1. 2017Attention Is All You Needcited by 59,932
  2. 2020Language Models are Few-Shot Learnerscited by 32,199
  3. 2023Improved Baselines with Visual Instruction Tuningcited by 4,362
  4. 2024Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolutioncited by 4,151
  5. 2024LLaVA-OneVision: Easy Visual Task Transfercited by 2,669
  6. 2016Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answeringcited by 2,498
  7. 2015VQA: Visual Question Answeringcited by 2,489
  8. 2023MMBench: Is Your Multi-modal Model an All-around Player?cited by 2,105
  9. 2023MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGIcited by 2,044
  10. 2024Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phonecited by 1,883
  11. 2022In-context Learning and Induction Headscited by 930
  12. 2023Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formattingcited by 778
28 more
  1. 2022A-OKVQA: A Benchmark for Visual Question Answering using World Knowledgecited by 702
  2. 2024MMMU-Pro: A More Robust Multi-discipline Multimodal Understanding Benchmarkcited by 494
  3. 2023V*: Guided Visual Search as a Core Mechanism in Multimodal LLMscited by 459
  4. 2023Large Language Models Are Not Robust Multiple Choice Selectorscited by 454
  5. 2024VLMEvalKit: An Open-Source Toolkit for Evaluating Large Multi-Modality Modelscited by 443
  6. 2023State of What Art? A Call for Multi-Prompt LLM Evaluationcited by 299
  7. 2024MME-RealWorld: Could Your Multimodal LLM Challenge High-Resolution Real-World Scenarios that are Difficult for Humans?cited by 284
  8. 2023Large Language Models Sensitivity to The Order of Options in Multiple-Choice Questionscited by 249
  9. 2022Leveraging Large Language Models for Multiple Choice Question Answeringcited by 199
  10. 2023Birth of a Transformer: A Memory Viewpointcited by 154
  11. 2024Divide, Conquer and Combine: A Training-Free Framework for High-Resolution Image Perception in Multimodal Large Language Modelscited by 151
  12. 2023LingoQA: Visual Question Answering for Autonomous Drivingcited by 144
  13. 2024When Benchmarks are Targets: Revealing the Sensitivity of Large Language Model Leaderboardscited by 134
  14. 2024Does Prompt Formatting Have Any Impact on LLM Performance?cited by 132
  15. 2021Between words and characters: A Brief History of Open-Vocabulary Modeling and Tokenization in NLPcited by 130
  16. 2024BetterBench: Assessing AI Benchmarks, Uncovering Issues, and Establishing Best Practicescited by 112
  17. 2024Efficient multi-prompt evaluation of LLMscited by 72
  18. 2025Can We Trust AI Benchmarks? An Interdisciplinary Review of Current Issues in AI Evaluationcited by 57
  19. 2025The Leaderboard Illusioncited by 57
  20. 2025Right Answer, Wrong Score: Uncovering the Inconsistencies of LLM Evaluation in Multiple-Choice Question Answeringcited by 27
  21. 2025FOCUS: Internal MLLM Representations for Efficient Fine-Grained Visual Question Answeringcited by 20
  22. 2024Order Matters: Exploring Order Sensitivity in Multimodal Large Language Modelscited by 19
  23. 2021A Cognitive Regularizer for Language Modelingcited by 11
  24. 2024WorldMedQA-V: a multilingual, multimodal medical examination dataset for multimodal language models evaluationcited by 11
  25. 2024Benchmark Data Repositories for Better Benchmarkingcited by 10
  26. 2025Set-LLM: A Permutation-Invariant LLMcited by 10
  27. 2025Selective Induction Heads: How Transformers Select Causal Structures In Contextcited by 7
  28. 2025Beyond Multiple Choice: Verifiable OpenQA for Robust Vision-Language RFTcited by 1

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Uncertainty Is Not a Safety Net for Clinical VQA, but Can It Anticipate Model Failure?cited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Robust Checkpoint Selection for Multimodal LLMs via Agentic Evaluation and Stability-Aware Ranking
  2. 2026Same Evidence, Different Answer: Auditing Order Sensitivity in Multimodal Large Language Models
  3. 2026Evaluating Stochastic Collapse and Implicit Bias in Multimodal Large Language Models
  4. 2024Multi-modal Preference Alignment Remedies Degradation of Visual Instruction Tuning on Language Models
  5. 2025Benchmarking and Mitigating MCQA Selection Bias of Large Vision-Language Models
  6. 2025Promptception: How Sensitive Are Large Multimodal Models to Prompts?
  7. 2026When Choices Become Risks: Safety Failures of Large Language Models under Multiple-Choice Constraints
  8. 2024MLLM-as-a-Judge: Assessing Multimodal LLM-as-a-Judge with Vision-Language Benchmark