Building with AI

Build your project with the latest AI models. Get better results faster.

Create uses AI to turns your prompt into a live site or app built with code.

Using Different Models

Each AI model has unique strengths and limitations. Some are higher quality but slower. Others trade off quality for speed. The same prompt made by a different model will get different results.

Model Overview

In a nutshell:

• Claude Sonnet 3.5: Best overall.

• GPT-4o (OpenAI): Sometimes better; definitely better at screenshots. Also good at fixing mistakes in Claude Sonnet generated code.

• Gemini 1.5 (Google): Slightly behind top models but depends on the task; good at long context.

• Haiku: Faster, but less high quality

• Opus: More detailed but takes longer.

• Groq w/ LLama 3.1: Fastest

Free vs. Pro Versions

• Free Version: Access to 1 model.

• Pro Version: Access to all models. You can update your generation settings from the top menu and select which model to use per slot. This means you can have, for example, four Claude Sonnet 3.5 models running simultaneously if you'd like.

Using Different Models

Create turns your spec into an app using all models. For every iteration, you can switch between them. It's like having 4 designers make your app at the same time.

Probabilistic outputs

If you run the same prompt through the same model multiple times, you get different results. As you add more details to your spec, you app gets more stable.

We offer two ways to generate to take advantage of this:

• Generate Button: Takes your old version and your new prompt to come up with a new version. Once you have a good starting point, use Generate to refine your output.

• Retry Button: Starts from scratch with just your spec. If you're not satisfied with the intial result, hit retry a few times to get different starting points.