OpenGCUA: Toward a Fully Open Stack for Generalist Computer Use
OpenGCUA develops an agent architecture that allows large language models to perform general computer use by autonomously deciding between visual interfaces and programmatic code execution. It integrates code execution and harness-to-model decision transfer, enabling agents to write unified programs across all interfaces, leading to significant performance gains on real devices. This means LLMs can become versatile computer operators rather than just text generators.