OpenGCUA: Toward a Fully Open Stack for Generalist Computer Use
OpenGCUA reports higher task scores across eight models, with 53% fewer reported steps on average than batched GUI-only execution.
On OSWorld-v2, the full configuration raised the mean partial score by 19.6 points. The agent uses code execution alongside visual interaction and transfers selected observation and task-state decisions from fixed rules to the model. The first release includes implementation code, prompts and evaluation trajectories for all eight models. A real-device harness and an SDK spanning computers, mobiles and browsers are preliminary releases.
HF Daily Papers' note
On OSWorld-v2, the full configuration raised the mean partial score by 19.6 points. The agent uses code execution alongside visual interaction and transfers selected observation and task-state decisions from fixed rules to the model. The first release includes implementation code, prompts and evaluation trajectories for all eight models. A real-device harness and an SDK spanning computers, mobiles and browsers are preliminary releases.
HF Daily Papers' note
score 5