CUA-Suite: Computer-Use Agent Video Dataset — Access Similar Capabilities via NexaAPI

A new dataset called CUA-Suite provides a large collection of human-annotated video demonstrations for training computer-use agents. Developers can access similar capabilities through the NexaAPI platform without requiring expensive GPU infrastructure.

Details

CUA-Suite is a significant advancement in computer-use agent (CUA) research, addressing the critical bottleneck of high-quality human demonstration data. The dataset includes ~55 hours and 6 million frames of expert video, with continuous 30fps screen recordings and multi-layered reasoning annotations. This preserves the full temporal dynamics of human interaction, going beyond previous datasets that only captured sparse screenshots. Models trained on CUA-Suite can automate complex desktop workflows, navigate GUIs without explicit programming, and understand multi-step task sequences from visual context. However, running these models locally requires expensive GPU infrastructure and complex setup. NexaAPI provides a solution, offering access to the vision and multimodal capabilities enabled by CUA-Suite at a cost of $0.003 per API call, without the need for GPU setup.