Cua Driver Bench evaluates computer-use tools on whether AI agents can complete real desktop tasks in the background without interrupting the user or relying on shell commands. Results from macOS and other platforms were produced with Scale AI's AgentEnv.