Imagine an AI that doesn't just chat, but actually navigates your phone, clicks through apps, and finishes complex tasks while you sleep. This is the promise of Qwen-UI-Agent, a new foundation model designed to operate reliably on real-world devices rather than just in isolated simulations. It represents a shift from passive assistants to active executors that can handle workflows across mobile, desktop, and web environments with minimal human intervention.

The core innovation lies in its ability to blend graphical user interface interactions with command-line execution. Unlike previous models that struggle with long-horizon tasks, this agent generates batched actions in a single turn, allowing it to plan and execute multi-step workflows efficiently. It uses a self-improving data flywheel where agents diagnose their own failures and construct new training environments, supported by online reinforcement learning that trains on trajectories exceeding one hundred turns. This setup enables the system to proactively initiate services and maintain stateful workflows across different platforms.

The results are striking, particularly on mobile devices. Qwen-UI-Agent sets a new state-of-the-art on mobile-use benchmarks, achieving an impressive 97.5% on AndroidDaily and 92.2% on MobileWorld-Real. It competes fiercely against frontier models like Opus 4.8, Gemini 3.1 Pro, and GPT-5.6 Sol, delivering competitive performance on computer and browser tasks as well. By combining diverse sandbox environments with a large-scale real-device runtime, the researchers have created an agent that bridges the gap between theoretical AI capabilities and practical, everyday utility.

This research marks a significant step toward general-purpose digital executors. By proving that agents can reliably operate on real devices and autonomously improve through reinforcement learning, the work suggests a future where AI seamlessly integrates into our digital lives, handling everything from simple app navigation to complex, multi-platform workflows with increasing autonomy and accuracy.