GUI agents have the potential to become a general purpose executor over existing digital devices. To advance them toward real-world use, we envision agents that operate reliably on real devices, execute workflows across platforms, combine GUI interaction with CLI execution, complete long-horizon tasks, proactively initiate useful services, and autonomously improve their capabilities with minimal human effort. Guided by this vision, we present Qwen-UI-Agent, a real-world centric foundation GUI agent spanning mobile, computer-use, web, and DeepSearch environments. 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 and generates batched actions in a single model turn. An AutoResearch-style data flywheel uses agents to construct tasks and environments, diagnose failures, and plan subsequent iterations. Online RL supports training on trajectories exceeding 100 turns, with over 10,000 concurrent environments accelerating rollout. A lightweight harness layer supports proactive service initiation and stateful workflows across mobile and computer. Across a broad suite of evaluations, Qwen-UI-Agent sets state-of-the-art performance on mobile-use benchmarks while delivering competitive performance on computer- and browser-use tasks against frontier models, including Opus 4.8, Gemini 3.1 Pro, and GPT-5.6 Sol. On mobile use, it achieves 82.1% on MobileWorld, 92.2% on MobileWorld-Real, and 97.5% on AndroidDaily. On computer use, it achieves 79.5% on OSWorld-Verified and a 40.0% partial-progress score on OSWorld-v2. On browser use and GUI grounding, it achieves 73.6% on WebArena and 81.5% on ScreenSpot-Pro, respectively.
核心贡献 · Key contributions
提出 Qwen-UI-Agent，一个覆盖移动端、电脑使用、网页与 DeepSearch 的面向真实世界的 GUI 智能体，并构建了由 100 多台真机组成的移动端运行时。 Introduces Qwen-UI-Agent, a real-world-centric GUI agent spanning mobile, computer-use, web, and DeepSearch, built on a 100+ device real mobile runtime.
提出统一动作空间，将 GUI 操作与 CLI 命令交织，并支持在单个模型回合内批量执行动作。 Proposes a unified action space that interleaves GUI operations with CLI commands and supports batched actions in a single model turn.
打造 AutoResearch 式数据飞轮，由智能体自动生成任务与环境、诊断失败并规划下一轮迭代。 Develops an AutoResearch-style data flywheel where agents generate tasks and environments, diagnose failures, and plan next iterations.
将验证器引导的在线强化学习扩展到超过 100 轮的轨迹，并利用超过 10,000 个并发沙箱环境加速 rollout。 Scales verifier-guided online reinforcement learning to trajectories beyond 100 turns using over 10,000 concurrent sandbox environments.
增加 harness 层，支持从移动通知触发主动服务，并在手机与电脑之间执行有状态的跨平台工作流。 Adds a harness layer enabling proactive service initiation from mobile notifications and stateful cross-platform workflows across mobile and computer.
在移动端任务上取得最先进结果，在 OSWorld、WebArena 和 ScreenSpot-Pro 上表现具有竞争力。 Sets state-of-the-art on mobile use and competitive results on OSWorld, WebArena, and ScreenSpot-Pro.
局限 · Limitations
真机移动端环境仅覆盖约 100 台设备和 150 款应用，可能无法代表长尾的设备与应用形态。 The real-device mobile setup covers about 100 devices and 150 apps, which may not represent the long tail of devices and applications.
跨平台工作流依赖 harness 进行编排与状态传递，端到端执行并非完全自主。 Cross-platform workflows depend on a harness for orchestration and state transfer, so end-to-end execution is not fully autonomous.
主动服务依赖移动通知作为信号，而通知往往不完整且有噪声，且重要操作仍需要用户确认。 Proactive service relies on mobile notifications as signals, which are partial and noisy, and consequential operations still need user confirmation.
验证器引导的在线强化学习依赖真实标准验证器，而许多真实任务仍难以自动合成此类验证器。 Verifier-guided online reinforcement learning requires ground-truth verifiers, which remain hard to synthesize for many real-world tasks.
评测主要基于现有基准，其更广泛的真实世界部署价值仍有待进一步验证。 Evaluation focuses mainly on existing benchmarks; broader real-world deployment utility remains to be demonstrated.