This paper introduces ParaGUIBench, which the authors describe as the first benchmark for parallel execution and coordination among multiple GUI agents. It includes a multi-device Docker infrastructure with a shared file system, 233 tasks across six categories, and efficiency metrics such as step reduction ratio and token cost. The proposed ParaGUI planner-worker system achieves a 46.4% success rate, 12.9 percentage points above the strongest serial baseline, Claude Sonnet 4.6, while using roughly half as many steps and less than half the tokens. The results target decomposable, long-horizon GUI workflows.
No heat snapshots are available in the last 24 hours.