VenusBench-GD is a dataset for the localization and understanding of graphical user interface (GUI) elements, released in 2025 by Ant Group in collaboration with iMean AI. Related research papers include... VenusBench-GD: A Comprehensive Multi-Platform GUI Benchmark for Diverse Grounding TasksThe aim is to evaluate the model's ability to accurately identify and locate target interface elements based on natural language instructions across different platform interfaces.

This dataset contains 6,166 manually labeled samples, covering two tasks: basic localization and advanced inference. Each sample consists of a screenshot of the interface and a corresponding natural language command. The data is built from 97 different applications and websites, covering web, mobile, and desktop platforms, and includes both Chinese and English interfaces. The basic tasks primarily assess the model's understanding of interface element types, text content, spatial relationships, and visual appearance. The advanced tasks further introduce inference, functional understanding, and the reasonable rejection of non-existent targets, placing higher demands on the model's global interface understanding and semantic inference capabilities. Through a multi-stage automated generation and manual review process, this dataset effectively reduces annotation noise and ambiguity while maintaining its scale, providing a reliable data foundation for evaluating GUI agents and multimodal models.