Thesys has released OUI-1, a diffusion model built on Google's DiffusionGemma 26B-A4B architecture that is designed specifically for generating user interfaces, scoring 71.7 percent on the Generative UI Benchmark against a 13 percent baseline from the underlying model.
OUI-1 generates OpenUI Lang in 256-token blocks, committing tokens progressively as confidence increases. With 4 billion active parameters from the 26 billion parameter base model, Thesys positions the release as optimised for the speed requirements of generative UI applications, where low latency is a practical constraint for production deployments.
The company also published benchmark results comparing OUI-1 and five other models across rendering success rates, token usage and cost under the same conditions. The benchmarks cover OpenUI, Google's A2UI and Vercel's json-render across 31 models. Results are available at openui.com/benchmarks.
Alongside OUI-1, Thesys released an Autofix API that addresses a known problem in UI generation: models generating valid but non-rendering output. In tests across 31 models, the Autofix API repaired 88 percent of broken OpenUI generations in a single call, which the company says reduces error states for end users without requiring full regeneration.
OUI-1 is the first model release from Thesys, which operates the OpenUI open standard for generative UI output.
To stay across the latest in cloud, AI and enterprise tech analysis from Compare the Cloud, subscribe to our weekly newsletter at https://www.comparethecloud.net/newsletter
