Alibaba's Tongyi-MAI team released OpenGCUA, an open-source stack that acts as a generalist computer-use agent. The system improves the mean partial-score of eight frontier models by 19.6 points on the OSWorld-v2 benchmark. It also reduces the number of steps required per task by 53.0 percent compared to a baseline. The release includes a macOS harness, a cross-device SDK, and evaluation trajectories for models like GPT-6 Astra and Claude Opus 5.
Read the full story at aiweekly.co.
Photo: fancycrave1 / Pixabay
