UI-Venus-2 is a general‑purpose GUI foundation agent that operates across mobile, web and desktop platforms through a unified closed‑loop reasoning‑action framework. To bridge the gap from benchmark models to real‑world deployment, we scale three critical dimensions in tandem: environment coverage expands to over 170 multilingual mobile apps and native desktop operating systems; task generation follows a deep‑research pipeline that produces function‑grounded instructions; verification introduces trace‑level and sample‑level evaluators, leveraging visual keypoints and multi‑model voting to supply reliable reinforcement‑learning signals. Safety‑aware mechanisms are also integrated to constrain consequential actions. By releasing an efficient, verifiable and self‑reflective open‑source foundation, UI-Venus-2 advances the development of generalizable GUI agents for practical applications.
Review