# OUI-1: A Gemma-based diffusion model for generative UI

> Thesys releases an experimental diffusion language model aimed at turning prompts into user interfaces, built atop Google's Gemma.

Published by The Open Weights on Sep 7, 2026. Canonical: https://theopenweights.com/news/oui-1-5lcn

## Key facts

- Company: Thesysdev
- Model: OUI-1
- Version: 1
- Category: Text / LLM
- Modalities: Code, Text / LLM
- License: Gemma (Open weights, commercial use allowed)
- Context window: 262K tokens
- Architecture: DiffusionGemmaForBlockDiffusion
- Precision: BF16
- Weights size: 51.6 GB
- Significance: minor
- Published: 2026-09-07
- Last verified: 2026-09-17
- Hugging Face: https://huggingface.co/thesysdev/OUI-1
- Canonical URL: https://theopenweights.com/news/oui-1-5lcn

A new model called **OUI-1** has appeared on Hugging Face, positioned as a diffusion language model for generative user-interface tasks. Published by the `thesysdev` account and built on Google's Gemma foundation, it is released under the Gemma license as an initial, minor release ([model card](https://huggingface.co/thesysdev/OUI-1)).

The premise is what makes it notable. Most text models generate tokens left to right; a diffusion approach instead refines an entire output in parallel across denoising steps. Pairing that technique with a UI-generation goal suggests an interest in producing structured layouts and code where the whole design is considered at once rather than assembled sequentially.

## What we know

- Base: Gemma, Google's open model family
- Type: diffusion language model, text and code
- Purpose: generative UI tasks
- License: Gemma

Several key details are not yet published, including parameter count, context length, and benchmark results. Without those figures it is hard to judge how OUI-1 compares to conventional code and UI generators, and buyers should treat it as an early experiment rather than a finished tool.

**Why it matters:** generative UI is a fast-growing niche, and applying diffusion to it is an unusual bet. If the approach holds up, parallel generation could offer a different speed-and-coherence tradeoff for building interfaces from natural language. For now, OUI-1 is worth watching as a signal of where open, Gemma-derived tooling is heading.

## Get the model

- [Hugging Face](https://huggingface.co/thesysdev/OUI-1)

## Sources

- [thesysdev/OUI-1](https://huggingface.co/thesysdev/OUI-1) — Hugging Face, Sep 7, 2026

---
Source: The Open Weights (https://theopenweights.com/). Aggregated and written by Claude, curated by humans. Cite as: "OUI-1: A Gemma-based diffusion model for generative UI", The Open Weights, Sep 7, 2026, https://theopenweights.com/news/oui-1-5lcn