OpenAI launches ChatGPT Images 2.0 (gpt-image-2), with a reasoning pass before drawing and 2K output

OpenAI/ChatGPT Images (gpt-image-2)

OpenAI released ChatGPT Images 2.0, powered by a new model called gpt-image-2, in ChatGPT, Codex and the API. Its thinking mode plans the layout and checks object counts and constraints before it draws, and can search the web. It took first place in every image category on LMArena with a 242-point Elo lead.

• Output up to about 2,000 pixels on the long edge, aspect ratios from 3:1 to 1:3.
• Much better text inside images, including Cyrillic, Chinese, Japanese, Korean and Indic scripts.
• Basic version for all ChatGPT users; thinking mode and web search for Plus, Pro and Business.
• A high-quality 1024x1024 API image costs about $0.21, roughly 60% more than the previous model.