Xiaomi presents Xiaomi-GUI-0, a native multimodal GUI agent trained and evaluated in a real-device closed loop. Physical mobile devices are the primary execution environment, with sandboxes used as auxiliary infrastructure. The system combines data for frequent tasks, long-tail intents, reflection, and memory, and uses an error-driven data flywheel to convert failures into corrected actions and recovery demonstrations. Training proceeds through supervised fine-tuning, step-level reinforcement learning, and agentic reinforcement learning. The report states 72.0% success on RealMobile and 78.9% on AndroidWorld.
No heat snapshots are available in the last 24 hours.