OpenUI OUI-1An open model built for Generative UI

Generate interfaces from your own components. OUI-1 writes OpenUI Lang, ready to render as it streams.

OpenUI OUI-1An open model built for Generative UI

Generate interfaces from your own components. OUI-1 writes OpenUI Lang, ready to render as it streams.

General-purpose models are built for text.
OUI-1 is purpose-built for interfaces. It transforms natural language into fast, reliable, interactive interfaces using your own components.

Streams as it generates

Screens start appearing before generation finishes.

Up to 500 tokens/sec

Ultra-fast generation for responsive interfaces.

Base model · 13%OUI-1 · 71.7%

71.7% benchmark score

5.5× its base model’s score on the Generative UI Benchmark.

4B active parameters

A compact model purpose-built for Generative UI.

Runs on consumer hardware

Bring interface generation onto your own hardware.

Build with OUI-1.

Read setup docs
  1. 1

    Run OUI-1

    Serve OUI-1 on your GPU with vLLM.

  2. 2

    Define your components

    Define your component library, then generate its system prompt.

  3. 3

    Render the interface

    Send a request to OUI-1 and stream its response into the Renderer.

Frequently asked questions

Have another question?
Join our Discord.

What is OUI-1?

OUI-1 is an open-weight model built for Generative UI. It is a finetune of Google's DiffusionGemma that turns a request into OpenUI Lang, which OpenUI renders into interactive components.

Why use it over a general-purpose model?

OUI-1 is trained to generate connected, valid interfaces. It scored 71.7% on the Generative UI Benchmark, compared with 13.0% for its base DiffusionGemma model, using the same benchmark protocol.

Can I use my own components?

Yes. Give OUI-1 your component signatures in the system prompt. It generates OpenUI Lang from that library, so the output can use your components and design system.

What hardware does it need?

OUI-1 has 26B total parameters, with 4B active per token. FP8 weights occupy about 25.8 GiB, plus runtime and cache memory. Check the model card for serving settings and plan GPU memory for your workload.

How do I connect it to OpenUI?

Serve OUI-1 with vLLM 0.24 or newer, generate a system prompt from your component library, and call its OpenAI-compatible endpoint. Render the streamed OpenUI Lang with an OpenUI renderer.

Can I use it commercially?

The weights are released under Apache 2.0. You can download, adapt, and deploy them on your own infrastructure, subject to the license terms.

Open-Source

Open weights, Apache 2.0

Download, customize, fine-tune, and deploy OUI-1 on your own infrastructure. Built for everyone, with no vendor lock-in.