Adept Fuyu-Heavy is a novel multimodal model designed specifically for digital agents. It excels in multi-modal reasoning, particularly in UI understanding, and performs well on traditional multimodal benchmark tests. Moreover, it demonstrates our ability to scale the Fuyu architecture and obtain all the associated benefits, including handling images of any size/shape and effectively reutilizing existing transformer optimizations. It also exhibits the capability to match or exceed the performance of models with the same computational level, although some capacity needs to be allocated for image modeling.