OpenAI launches GPT-4o CodeDesigner: a multimodal model tailored for UX engineers

AI · 5 min read

OpenAI today unveiled GPT-4o CodeDesigner, a specialized branch of its GPT-4o family that targets design-to-code workflows for UX engineers and product teams. The model supports image, sketch, and AR input and outputs deterministic, framework-aware component code (React, SwiftUI, Compose) along with accessibility annotations and tokenized design tokens.

CodeDesigner introduces a new 'spec mode' that emits machine-readable design system artifacts (tokens.json, component manifests, and test scaffolds) rather than freeform suggestions. OpenAI says the model was trained with a mix of licensed UI datasets and partnered design system repositories and includes a safeguards layer to avoid leaking proprietary component code during collaborative sessions.

The new product ships with plugins for Figma, VS Code, and Xcode that allow designers to request live code previews, visual diffs, and auto-generated accessibility reports. Pricing is API-based with a studio tier that bundles private fine-tuning and on-prem model enclaves for enterprise customers, targeting design ops teams that need reproducible component outputs.