OpenAI releases Mosaic-1 — a multimodal model tuned for product design

AI · 6 min read

Mosaic-1 is positioned as a design-focused multimodal model combining text, image, and layout understanding. It ships with a 'Design Mode' that returns wireframes, CSS suggestions, and accessibility annotations in a single response, streamlining the early product design loop.

The model was trained on a curated corpus of design systems, component libraries, and WCAG-compliant examples to reduce hallucinations on technical specifications. It also supports a new s-expression schema for returning structured component hierarchies designers can drop directly into tools.

OpenAI is offering Mosaic-1 via an API tier with expanded context windows and a partnership program for design platforms to embed the model natively. Early partners include a range of prototyping and documentation platforms seeking to automate repetitive drafting tasks.