CUA-Universe: A Scalable and Dynamic Environment for Hybrid GUI+CLI Agents

The researchers introduced CUA-Universe, a scalable and dynamic hybrid graphical user interface (GUI) and command-line interface (CLI) proxy environment. This environment uses three core components: App-Forge, Task-Weave, and Path-Steer to transform real desktop software into reproducible virtual machine environments that include visual state checks and command-line operations. The 9B model trained on this data performed excellently in the CUA-Verse, OSWorld, and OSWorld-MCP benchmarks, improving task success rates or scores respectively, and significantly reducing the number of steps and tokens required. This achievement represents a shift from inefficient graphical interactions to efficient hybrid orchestration.