Choosing between GPT Image 2.5 and DALL·E 3 reads like a two-model comparison. OpenAI’s own documentation frames it differently: DALL·E is no longer offered as a parallel option inside ChatGPT, so the reader’s real question is how far that retirement reaches.
Everything below comes from OpenAI’s release notes and help centre. There are no measured scores and no per-image prices, because OpenAI publishes neither.
Start from the task
- Working inside ChatGPT and editing an image by talking about it: the current generation documents that workflow in detail
- Producing an image and managing it afterwards: images are saved automatically under Images, on the same surface both generations lived in
- Wanting a documented answer to “is DALL·E still an option”: the help centre answers it directly, and the answer is no
- Checking whether an existing integration still works: that belongs to OpenAI’s own model and deprecation pages, not to a comparison article
The question is not which of the two to pick. It is what a DALL·E habit has to become.
Align the versions before comparing
Most pages ranking for this term compare the wrong generation.
- GPT Image 2.5 is the current OpenAI generation, released on 8 September 2026, and it ships in the API as two variants:
GPT-Image-2.5 Flare, the default for most applications, andGPT-Image-2.5 Sunburst, for premium visual workflows that need tighter control across edits - DALL·E 3 is the earlier generation, documented in OpenAI’s help centre only as something that has been retired from ChatGPT
- Pages titled “GPT Image 2 vs DALL-E 3” compare the previous generation, and later pages often reuse those older findings without re-checking them
A claim about GPT Image 2 is not a claim about GPT Image 2.5.
What OpenAI documents for each generation
This asymmetry is worth understanding before you read any comparison.
Documented for GPT Image 2.5:
- Image generation latency reduced by up to 50% compared with Images 2.0
- Sketch, for drawing a reference in the message box; templates for common formats such as flyers and product photos; comments placed on an image; and prompt sharing
- Two API variants, with Flare documented at 50% lower latency than GPT-Image-2
Documented for DALL·E:
- A retirement notice in the help centre, directing users to ChatGPT Images
- A related note that GPTs with Image Generation enabled in Capabilities create images with the new generation model
That is the whole of it. OpenAI’s current image documentation carries no DALL·E capability list, so a detailed DALL·E 3 feature table published this month did not come from OpenAI.
Editing becomes a conversation
The larger change is not a better first image. It is that editing stops being a single pass.
- Two documented ways to edit: select part of the image with the selection tool and describe the change in chat, or skip the selection tool and describe the edit in the conversation panel
- The editor offers Select, Aspect ratio, Undo, Redo, Cancel and Save; on mobile it also offers Resize and Remove
- You can generate in any aspect ratio, chosen with the picker or stated in the prompt
- Comments placed directly on an image narrow the scope of a change
- OpenAI states the limit as plainly as the capability: highlights are not always precise, and edits may extend beyond the area you selected
A documented caveat is more useful to plan around than a promise.
Text rendering and typography
Both generations get discussed online in terms of text inside images. Only one of them is still documented.
- The release lists sharper details and more precise editing among its improvements
- ChatGPT Images is documented as able to add text, add details within the image, or make the background transparent
- Work that leans on lettering — signage, labels, another script — we cover separately in text rendering in GPT Image 2.5
- No accuracy figure for text appears on either page, so a numeric verdict is not available from official material
What the retirement note actually covers
Most pages flatten two different statements into one.
- The help centre answers the DALL·E question in plain language: the official DALL·E GPT has been retired from ChatGPT, and ChatGPT Images is what replaces it
- That sentence is about the product surface inside ChatGPT. It is not a statement about an API model identifier
- The pages referenced here publish no
dall-e-3sunset date, no migration deadline and no deprecation window - Pages that quote a specific retirement date for the API model are quoting someone other than OpenAI; treat the date as unverified until it appears on OpenAI’s own model or deprecation page
The safe reading is narrow: inside ChatGPT the DALL·E option is gone, so a chat-based habit has to move. Whether an automated integration is affected is a separate question with a separate answer.
What neither page publishes
Comparisons usually fill these gaps with numbers. This one will not.
- No per-image price: the release links to a separate pricing page instead
- No quota, allowance or rate limit for the current generation
- No benchmark comparing the two generations, and no accuracy figure for either
- No published cap on resolution, output format list or reference-image count
If a page puts a dollar figure next to a model name today, check whether OpenAI published it. In the pages we used, it did not. For the two API variants themselves, see the GPT Image 2.5 API overview.
A comparison you can repeat
Because the official record is thin on numbers, compare the parts that are documented, on your own brief.
Run the same brief twice and watch how many turns each one needs: “Design a poster for a weekend market. Keep the headline legible, keep the palette to two colours, and leave the bottom third for opening hours.”
| Check | What OpenAI documents | What stays unpublished |
|---|---|---|
| Can you edit by describing it? | Yes — selection tool or plain description in chat | No accuracy or retry figure |
| Can you fix one region? | Comments on the image, with a stated imprecision | No measured precision rate |
| Can you set the frame? | Any aspect ratio, via picker or prompt | No published size list |
| Does one pass usually suffice? | Not stated either way | No published attempt count |
Keep the brief identical on both sides, or the test measures your prompting rather than the model. For the wider capability picture, see the GPT Image 2.5 overview.
Frequently asked questions
Is DALL·E 3 still available in ChatGPT?
OpenAI’s help centre states that the official DALL·E GPT has been retired from ChatGPT and that users should use ChatGPT Images to create or edit images.
Does that mean the DALL·E API is gone too?
The retirement note is about the ChatGPT surface. The pages referenced here publish no API sunset date, so confirm the model’s status on OpenAI’s own model and deprecation pages before assuming either answer.
Should I rewrite old DALL·E prompts for GPT Image 2.5?
There is no official migration guide for prompts, and no published list of prompt differences. Rewrite where a habit is documented to behave differently — for example, describing an edit instead of re-describing the whole image.
Can I share the prompt that produced an image?
Yes. Prompt sharing is documented: when you share an image, you can also choose to include the prompt that produced it, which we cover in prompt sharing in GPT Image 2.5.