ChartAct: A Benchmark for Dynamic Chart Understanding

A new benchmark called ChartAct tests how well AI models understand dynamic, interactive charts—the kind found on real websites where key data appears only after hovering, clicking, or zooming, according to a paper posted to arXiv. [1] Existing chart-understanding benchmarks focus on static images, but real-world charts are often dynamic and interactive. Key information may only appear after actions such as hovering, clicking, zooming, or dragging. Dynamic chart understanding therefore requires models to identify visible content, choose proper interactions, and reason over changing chart states. [2] To evaluate this ability, researchers constructed ChartAct, an interactive benchmark that collects and filters 673 dynamic charts from 8 real chart websites, covers 7 common chart types, and builds 1,440 high-quality question-answer samples. [1][2] Each sample is instantiated in two environments—Dynamic Chart and Dashboard Chart—to evaluate understanding under different contexts. [2] The team systematically evaluated 11 advanced multimodal models and GUI agents. [1] The strongest model, Claude-Opus-4.7, achieved an average success rate of 84.5%, while most models remained below 60%. [1][2] The paper also includes detailed failure attribution and case analysis. [2] The benchmark arrives as technology companies such as OpenAI, Google, xAI, and Meta pursue artificial general intelligence, a hypothetical type of AI that would match or surpass human capabilities across virtually all cognitive tasks. [4] A 2020 survey identified 72 active AGI research and development projects across 37 countries. [4] Chart understanding sits within the narrower domain of artificial narrow intelligence, where competence is confined to well-defined tasks, though the ability to reason over changing visual states is considered a step toward more general capabilities. [4] The ChartAct code is publicly available on GitHub. [2]