LLaDA-UI ships a 16.7B GUI agent

InclusionAI released LLaDA-UI, an open 16.7B-parameter mixture-of-experts diffusion model that turns screenshots into structured GUI actions. Its model card lists a roughly 32 GB BF16 checkpoint, native-resolution vision, and support for mobile, desktop, and web interfaces. Builders evaluating local computer-use agents should treat 16.7B as a weight count rather than a production serving budget, and compare its action loop with Google’s open diffusion model before granting click permissions.