During the World Artificial Intelligence Conference (WAIC), Honor officially launched its self-developed multimodal perception large model - the MagicGUI open-source large model. As Honor’s first open-source GUI large model, MagicGUI, with a 7B (7 billion) parameter scale, empowers underlying AI agents with multimodal perception and autonomous execution planning capabilities. It has achieved industry-leading performance in mobile-end visual positioning, page comprehension and Q&A, single-step operation execution, and action sequence decomposition and execution.