OpenAI retired the DALL-E GPT from ChatGPT on Sunday. That move ends the last visible trace of the company’s earlier image generation system inside the chatbot. The dedicated tool, once available through ChatGPT’s model picker, has been replaced by ChatGPT Images. The new capability runs on OpenAI’s gpt-image-1 and gpt-image-1-mini models.

The change first appeared in ChatGPT’s release notes on July 31. It now gives every user—including those on free accounts—access to the upgraded image generation tool on both web and mobile platforms.

What Changes for Users

ChatGPT Images handles complex scene descriptions with greater accuracy than DALL-E. It produces fewer visual artifacts and does a better job interpreting compositional requests, such as specific lighting conditions or spatial relationships between objects. The tool is now the default for image generation across all account tiers.

One feature stays behind a paywall. “Images with thinking,” which applies reasoning steps before generating an image, is available only to Plus, Pro, and Business subscribers. Enterprise and Edu access will follow. Custom GPTs that already have image generation enabled are unaffected by the shutdown.

Download Your Images Now

Any images created through the DALL-E GPT exist only inside the conversations where they were generated. OpenAI has urged users to download anything they want to keep. Those images will disappear when the associated conversations are deleted.

Users can find older images through ChatGPT’s search function and save them via the browser or the mobile app.

Part of a Broader Consolidation

The retirement fits a pattern OpenAI has followed throughout 2026. The company removed DALL-E 2 and DALL-E 3 from its developer API on May 12, after issuing a deprecation notice in November 2025. Earlier in August, OpenAI retired its o3 reasoning model from ChatGPT after a ninety-day sunset period.

Further model consolidation is already scheduled. In June, OpenAI announced that gpt-image-1-mini, gpt-image-1.5, and chatgpt-image-latest will all be retired on December 1 and replaced by gpt-image-2.

The approach is straightforward: fewer specialized entry points, each doing more work, instead of a growing list of tools that each demand their own learning curve.