The Qwen-UI-Agent technical report introduces a foundation GUI agent designed for real-world digital environments, including mobile, computer-use, web, and DeepSearch platforms. To bridge the gap between simulation and real-world utility, the researchers developed a unified action space that interleaves graphical user interface (GUI) operations with command-line interface (CLI) execution, allowing the model to generate batched actions in a single turn. The training methodology utilizes an "AutoResearch-style" data flywheel, where agents autonomously construct tasks, environments, and failure diagnostics to drive iterative improvement. Furthermore, the framework employs online reinforcement learning (RL) capable of handling trajectories exceeding 100 turns, supported by over 10,000 concurrent environments to accelerate rollouts. Empirically, Qwen-UI-Agent demonstrates state-of-the-art performance on mobile benchmarks, specifically achieving 82.1% on MobileWorld, 92.2% on MobileWorld-Real, and 97.5% on AndroidDaily. In computer-use tasks, it reached 79.5% on OSWorld-Verified and a 40.0% partial-progress score on OSWorld-v2. Additionally, the agent shows strong capabilities in browser and grounding tasks, scoring 73.6% on WebArena and 81.5% on ScreenSpot-Pro, remaining competitive against frontier models such as Opus 4.8, Gemini 3.1 Pro, and GPT-5.6 Sol.