The new DALL-E? What changes with GPT Image 2.5, Sunburst and Flare

From asking for an image to fixing the one you have

You want a bookshop poster and like the composition, but the date is wrong. Asking for everything again is not always the most useful option: you need to correct one part. OpenAI’s current documentation includes generation and editing with GPT Image 2.5, in two variants called Sunburst and Flare. Although DALL-E may be the name you remember, distinguish it from the model a particular service uses. This explains what appears in the guides, rather than reporting a launch today.[1]

Sunburst and Flare: deciding what matters in your image

OpenAI presents Flare as the speed-oriented option and Sunburst as the quality-oriented one. Sunburst’s listing identifies a version dated 8 September 2026. These are provider descriptions; we have not measured which responds sooner or how much an image improves. For our imaginary poster, the question is which part of the job we are doing: trying several compositions or refining the chosen one. One name is not necessarily the right answer for every assignment.[2][3]

The bookshop example: an instruction you can actually check

Imagine a wooden shopfront, a table of books and a sign reading “Talk on Friday”. “Change Friday to Saturday and preserve the rest” is a more concrete request than “make a better poster”. You can check the result against a clear intention: inspect the word, then the books, shadows and shopfront. This illustrates a workflow; we are not showing an image produced by those models or promising error-free execution. Nor do we claim this version invented image editing.

An edit can also change what you wanted to keep

OpenAI’s guide warns that repeated edits can change details intended to stay unchanged. A correct date is therefore not enough to accept the whole picture. In our example, is the shop’s name still legible? Does any lettering look convincing until you actually read it? Keeping the previous version lets you compare and reject an edit that fixes one thing while breaking another. The useful result is an image you can use, not a collection of unchecked variations.[2]

DALL-E in Azure and GPT Image in the API are not your chat plan

Microsoft Learn dates the retirement of DALL-E 3 in Azure to 4 March 2026. That information concerns Azure; it does not establish a universal retirement of every tool called DALL-E. The linked OpenAI guides describe access for integrating images into applications through an API. They do not establish what your ChatGPT subscription includes. If you simply want to make a poster in a chat, look at that app’s options; if you are building a tool, check its documented model and access conditions.[4][1]