OpenAI: ChatGPT Images 2.0 launches with reasoning, arrives in Figma same day

OpenAI launched ChatGPT Images 2.0 on April 21, 2026 — the first image model the company has built with native reasoning built into the generation process. The model, also available via API as gpt-image-2, thinks before it draws: it runs a reasoning step over the prompt before generating an image rather than generating directly from the text input.

The practical differences over its predecessor are in accuracy and consistency. The model renders text in images with high fidelity across more than a dozen scripts, including Japanese, Arabic, Korean, Devanagari, Cyrillic, Chinese, Bengali, and Greek — addressing one of the longstanding weaknesses of generative image tools. It supports batches of up to eight images with object and character continuity across the set, and outputs up to 2K resolution with aspect ratios from 3:1 to 1:3.

Figma integrated the model on launch day, making it available inside Make Image and Edit Image across Figma Design, Draw, Slides, Buzz, FigJam, and Weave. For designers already working in Figma, this means generating reference images without leaving the application. The model’s stronger text rendering makes it more practical for interface mock data, multilingual UI assets, and any design work where placeholder text in the image needs to be readable rather than obviously artificial.

OpenAI retired DALL-E 2 and DALL-E 3 on May 12, 2026.

For product designers, the most relevant change is the combination of in-context reasoning and Figma integration. The model produces more reliable outputs from complex prompts — “a notification banner in Japanese showing a booking confirmation with an icon on the left” — than was possible with previous tools. Whether it meaningfully changes design workflows depends on how much time a designer currently spends generating or sourcing reference assets, but for teams working on multilingual products, the text rendering alone represents a real improvement over available alternatives.