OpenAI launches GPT-4o Studio — on-device multimodal tuned for designers

AI · 5 min read

OpenAI today announced GPT-4o Studio, a new variant of the GPT-4o family optimized for multimodal interactions and low-latency on-device inference. The model focuses on sketch-to-vector translation, layout recommendations, and contextual copy generation that understands design systems and component libraries.

The launch bundles a Figma plugin, a macOS SDK using Apple's ML Compute, and an iPad runtime with quantized weights to run locally with limited memory. OpenAI highlighted privacy-by-default: artboards and sketches can be processed entirely on-device, with optional cloud fallbacks for heavier tasks like photorealistic rendering.

Designers who tested early builds praised the rapid layout iterations and contextual suggestions tied to design tokens and component libraries, though some noted edge cases around iconography and brand-color fidelity. OpenAI said partners will get extended APIs for asset export in SVG, XD, and code snippets for React and SwiftUI.