Qwen UI agent reaches 82.1% on MobileWorld

Qwen-UI-Agent reached 82.1% on MobileWorld, 92.2% on MobileWorld-Real, and 97.5% on AndroidDaily while using one action space across phones, computers, browsers, and command-line tools. The technical report also reports 79.5% on OSWorld-Verified and says online reinforcement learning ran trajectories longer than 100 turns across more than 10,000 concurrent environments, although the results remain author-reported preprint scores. Teams tracking the economics of computer-use agents should file the cross-platform design, then test real workflows: a unified harness can reduce integration seams, but benchmark breadth does not prove reliable long-horizon execution.