Cua-Bench – Benchmarks & RL Environments for AI Agents in GUI Desktop Automation
Summary
Cua-Bench provides benchmarks and RL environments for computer-use agents that operate GUI desktops across OSWorld, ScreenSpot, and Windows Arena, with exportable trajectories for training. It is part of an open-source platform for building, benchmarking, and deploying agents that can control full desktops, making it relevant for AI tooling, automation research, and enterprise-scale automation experiments.