#Tag (1)

CUA-Master: Scaling Computer-use Agents through Open-World Recursive Improvements

CUA-Master is a nested double-loop recursive improvement framework that mines authentic workflows from instructional videos, reconstructs executable trajectories, propagates seed tasks into verifiable synthesized tasks, and uses feedback-guided snapshot rollback retries to collect reliable computer-use trajectories. Its 27B model scores 80.2 on OSWorld Verified, 36.9 on OSWorld 2.0, 43.9 on WeaveBench, and 25.0 on Agents' Last Exam.