Hugging Face has released a technical report introducing Qwen-UI-Agent, a foundation GUI agent designed for real-world applications. The agent spans mobile, computer-use, web, and DeepSearch environments, aiming to operate reliably on actual devices and execute workflows across platforms.
Qwen-UI-Agent combines diverse sandbox environments with a large-scale real-device mobile runtime. Its unified action space interleaves GUI operations with CLI execution, generating batched actions in a single model turn. This design supports long-horizon tasks and proactive service initiation, with a lightweight harness layer enabling stateful workflows across mobile and computer.
The training pipeline leverages an AutoResearch-style data flywheel, where agents construct tasks, diagnose failures, and plan iterations. Online reinforcement learning supports trajectories exceeding 100 turns, accelerated by over 10,000 concurrent environments.
In evaluations, Qwen-UI-Agent achieves state-of-the-art performance on mobile-use benchmarks, including 82.1% on MobileWorld, 92.2% on MobileWorld-Real, and 97.5% on AndroidDaily. On computer use, it scores 79.5% on OSWorld-Verified and a 40.0% partial-progress score on OSWorld-v2. For browser and GUI grounding, it reaches 73.6% on WebArena and 81.5% on ScreenSpot-Pro, respectively.
The agent remains competitive with frontier models such as Opus 4.8, Gemini 3.1 Pro, and GPT-5.6 Sol on computer and browser tasks. The project page and code are publicly available.