aiqbench analysis
This article introduces DRAGON, a new benchmark designed to evaluate whether vision-language models genuinely ground their reasoning in specific visual evidence within diagrams rather than relying on textual shortcuts. By requiring models to identify bounding boxes for supporting elements like labels and connectors, the study highlights significant gaps in current VLMs' interpretability across scientific charts, circuits, and maps. This work is critical for advancing reliable multimodal AI systems that can rigorously analyze structured visual data.
DRAGON: A Benchmark for Evidence-Grounded Visual Reasoning over Diagrams
arXiv:2604.25231v2 Announce Type: replace-cross Abstract: Diagram question answering (DQA) requires models to interpret structured visual representations such as charts, maps, infographics, circuit schematics, and scientific diagrams. Recent vision-language models (VLMs) often achieve high answer accuracy on these tasks, yet correct answers do not guarantee that models ground their reasoning in the diagram regions that support the prediction. Models may instead rely on textual correlations or dataset artifacts without identifying the visual evidence required to verify the answer. This limitation prevents reliable evaluation of diagram reasoning and reduces interpretability. We introduce DRAGON, a benchmark for evaluating evidence-grounded visual reasoning in diagrams. Given a diagram, a question, and the correct answer, a model must predict bounding boxes that correspond to the visual elements required to justify the answer. These evidence regions may include answer-bearing components, textual labels, legends, axes, connectors, and other supporting structures involved in the reasoning process. The DRAGON dataset contains 11,664 annotated question instances from six diagram QA datasets: ChartQA, Circuit-VQA, InfographicsVQA, MapIQ, MapWise, and AI2D, with a 2,445-instance test set carrying human-verified evidence annotations and a standardized evaluation framework. Our evaluation on recent VLMs reveals that even Claude Opus 4.6 achieves a Grounding IoU of only 23.8% on AI2D, showing that faithful visual grounding remains an open challenge. DRAGON supports future research on models that ground their predictions in visual evidence.
Abstract:Diagram question answering (DQA) requires models to interpret structured visual representations such as charts, maps, infographics, circuit schematics, and scientific diagrams. Recent vision-language models (VLMs) often achieve high answer accuracy on these tasks, yet correct answers do not guarantee that models ground their reasoning in the diagram regions that support the prediction. Models may instead rely on textual correlations or dataset artifacts without identifying the visual evidence required to verify the answer. This limitation prevents reliable evaluation of diagram reasoning and reduces interpretability. We introduce DRAGON, a benchmark for evaluating evidence-grounded visual reasoning in diagrams. Given a diagram, a question, and the correct answer, a model must predict bounding boxes that correspond to the visual elements required to justify the answer. These evidence regions may include answer-bearing components, textual labels, legends, axes, connectors, and other supporting structures involved in the reasoning process. The DRAGON dataset contains 11,664 annotated question instances from six diagram QA datasets: ChartQA, Circuit-VQA, InfographicsVQA, MapIQ, MapWise, and AI2D, with a 2,445-instance test set carrying human-verified evidence annotations and a standardized evaluation framework. Our evaluation on recent VLMs reveals that even Claude Opus 4.6 achieves a Grounding IoU of only 23.8% on AI2D, showing that faithful visual grounding remains an open challenge. DRAGON supports future research on models that ground their predictions in visual evidence.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
arXiv:2604.25231 [cs.CV]
(or arXiv:2604.25231v2 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2604.25231 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 320 words
Related coverage
PhysAlign: A Benchmark for Evidence-Grounded Role Alignment in Multimodal Physics Reasoning
arXiv · Sep 29, 2026
VistaHop: Benchmarking Multi-hop Visual Reasoning for Visual DeepSearch
arXiv · Jun 3, 2026
Article: The AI Productivity Paradox in Test Automation: Moving beyond Structural Validation to Perception and Intent
InfoQ · Jun 1, 2026
Thinking with Visual Grounding
arXiv · Jun 16, 2026