Small AI Models Can Now Design App Screens for a Fraction of the Cost

Designing an app's interface—the buttons, forms, and screens you tap every day—is slow, careful work. Big AI models can now generate this code automatically, but they're expensive and can be slow for everyday use. A new study shows that much smaller, cheaper AI models can do nearly the same job, at a fraction of the cost. That's potential good news for any company that builds apps, and for you if you use them.

The secret is a change in how the AI works. Instead of asking the model to write interface code from scratch, researchers used a system called A2UI, which lets the AI pick from a catalog of pre-built components and connect them to data. Think of it like assembling a room from standard furniture rather than carving every piece by hand. This approach is safer and more consistent for real products—and it turns out smaller models can handle it surprisingly well.

In the study, a model with 4 billion parameters—small enough to run cheaply and quickly—captured about 98% of the quality of a much larger frontier model, while reducing costs by more than tenfold. The researchers also found that giving small models a large catalog of components helps, and that how you prepare training examples matters a lot. Those are practical lessons for developers trying to ship interfaces without breaking the bank.

The catch? It's not perfect. The small model still had a 2–3% quality gap, so visual polish and tricky edge cases may need human oversight. But for many everyday screens and forms, cheap and fast might be good enough. That could let smaller teams build quality interfaces, speed up updates, and keep costs down—benefits that often end up in your hands.