OUI-1: world's first model for Generative UI (www.openui.com)

OUI-1 has been announced as the world's first model designed specifically for Generative User Interfaces, utilizing a finetuned version of the DiffusionGemma model. This innovative model can create user interfaces in openui-lang on consumer-grade GPUs, such as the RTX 5090. OUI-1 addresses the critical constraints of speed, reliability, and local execution, aiming to make agent-driven interfaces accessible on standard hardware. The incorporation of OpenUI Lang, which is more efficient than traditional formats, allows for nearly immediate interface generation, greatly enhancing user experience. Significantly, OUI-1 achieved a score of 71.7% on the Generative UI Benchmark, a substantial improvement from DiffusionGemma’s initial 13.0%. This leap in performance was realized through a two-stage training process that combined supervised fine-tuning and self-distillation, enabling the model to both generate high-quality interfaces and maintain speed. Notably, OUI-1 surpassed other models with higher parameter counts, proving that efficiency and reliability can coexist even in smaller models. The eventual goal is to ensure interfaces can be generated reliably in under a second, thereby transforming how software is developed and utilized in consumer environments.