DALL·E 3 was OpenAI's text-to-image model. OpenAI shut down the DALL·E 2 and DALL·E 3 API endpoints on 12 May 2026 and now directs users to its GPT Image models. This page explains what happened and what replaced it.
DALL·E 3 is no longer available. OpenAI announced its deprecation on 14 November 2025 and shut down the DALL·E 2 and DALL·E 3 API endpoints on 12 May 2026.
This is not a rebrand or a quiet repositioning. The endpoints were switched off, and OpenAI's own deprecations page now directs developers to its GPT Image models instead.
Because DALL·E 3 is still heavily searched and widely referenced in older articles, this page sets out the timeline, what replaced it, and what to do if you built something on it.
| Current status | Retired — no longer available |
|---|---|
| Shutdown date | 12 May 2026, for both dall-e-2 and dall-e-3 API endpoints |
| Announced | 14 November 2025 |
| Replacement models | gpt-image-2, gpt-image-1, and gpt-image-1-mini |
| In ChatGPT | Replaced by GPT Image 1.5 in December 2025, then GPT Image 2 in April 2026 |
| Consumer name | ChatGPT Images 2.0 |
| Also being retired | gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest, from 1 December 2026 |
| Source | OpenAI's published API deprecations page |
The retirement ran over about eighteen months, and knowing the order of events explains why so many articles contradict each other.
DALL·E 3's real advance was prompt adherence. Earlier image models produced attractive pictures that ignored half the instruction; DALL·E 3 followed complex prompts closely enough that you could describe a scene and get that scene.
It also handled text inside images far better than its predecessors, which was the practical blocker for anything involving a sign, a label, or a title.
Its integration into ChatGPT mattered as much as the model. Being able to describe an image conversationally, then refine it in the same thread, brought image generation to people who would never have opened a specialist tool.
OpenAI's recommended replacements are gpt-image-2, gpt-image-1, and gpt-image-1-mini. GPT Image 2 is the current flagship and the one OpenAI points new work toward.
GPT Image 2 is a different kind of model rather than a bigger DALL·E. It is described as OpenAI's first image model with native reasoning — able to search the web during generation, check its own output, and produce several coherent images from one prompt.
In ChatGPT the change happened without users needing to do anything. Anyone still typing image prompts into ChatGPT has been on GPT Image for some time, whether or not they noticed.
Calls to the DALL·E endpoints now fail rather than degrade, so any integration still pointing at them is broken, not merely outdated.
The migration path is to gpt-image-2. Note that two of the interim replacements — gpt-image-1-mini and gpt-image-1.5, along with chatgpt-image-latest — are themselves scheduled for shutdown on 1 December 2026, so migrating to those buys only a short reprieve.
The lesson worth taking from this cycle is to pin image generation behind a thin internal interface. OpenAI has retired image models three times in roughly a year, and nothing suggests that pace is finished.
The shift is architectural, not incremental. DALL·E 3 took a prompt and produced a picture in one pass; GPT Image 2 reasons about the request before and during generation.
In practice that means it can look something up mid-generation rather than guessing at a detail, inspect what it produced against the instruction, and return a set of images that belong together instead of variations that drift apart.
For anyone who used DALL·E 3 heavily, the biggest day-to-day difference is consistency across a batch. Producing several images that share a style and hold to one brief was the weakest part of the old workflow.
Image model retirements raise a question people rarely think about until it matters: what happens to images you already generated and shipped.
Work produced under a model that no longer exists does not become unusable, but reproducing it does. If a campaign depends on being able to regenerate assets in a consistent style, that capability leaves with the model.
Where visual consistency matters over years rather than weeks, keep the source files and the exact prompts, and treat any specific model as temporary infrastructure.
DALL·E 3 was the best-known image model of its generation, so an enormous volume of tutorials, comparisons, and reviews were written about it. Almost none have been updated.
Search results therefore present a model that has not existed since May 2026 as a current option, often with confident feature comparisons against tools that are still running. Some of those pages continue to rank well, which keeps the misinformation circulating long after the shutdown.
Treat any undated DALL·E 3 article as historical. OpenAI's own deprecations page is the authoritative record, and it is unambiguous about both the announcement and the shutdown date.
If you came here looking for DALL·E 3 specifically, the closest continuation is GPT Image 2 — same lineage, same ChatGPT integration, stronger prompt handling.
If you were choosing between DALL·E 3 and alternatives, that comparison needs redoing from scratch. The competitive landscape moved substantially while DALL·E 3 was being wound down, and a two-year-old comparison tells you nothing useful.
Weigh prompt adherence, text rendering, licensing terms for commercial use, and whether you need generation inside an existing workflow or as a standalone tool.
Add one criterion this episode makes obvious: how the vendor handles retirement. A provider that publishes dated deprecation notices well in advance is easier to build on than one that changes a model underneath you without saying so.
DALL·E 3 was an important model that is no longer available. It brought reliable prompt adherence and usable in-image text to a mass audience through ChatGPT, and it shaped what people expect from image generation.
Its retirement is a straightforward product decision rather than a failure — OpenAI replaced it with something more capable and consolidated onto one image line. Judged on influence rather than availability, it did more than any other model to make image generation ordinary.
For current capabilities, plans, and limits, go to OpenAI directly. Anything describing DALL·E 3 as a live option is describing the past.
Historic: strong prompt adherence for complex, multi-element descriptions
Historic: markedly better text rendering inside generated images
Historic: conversational generation and refinement inside ChatGPT
Historic: available through both ChatGPT and the OpenAI API
Current: superseded by GPT Image 2, OpenAI's recommended replacement
DALL-E 3 has an AInexfinder Editor Score of 3.8/5.
AInexfinder does not list plan prices or billing details. For current pricing, plans, and trials, visit the official DALL-E 3 website.
Visit official website for pricingVendor pricing, credits, and billing policies change over time. Always confirm on the official site before you buy.
Log in to write a review.
No reviews yet. Be the first to share your experience with DALL-E 3.
Assigned reviewer
Daniel ReedSenior AI Tools Reviewer
Daniel reviews AI tools the slow way — by actually using them on real projects. His reviews cover what works, what breaks, and who each tool is genuinely a good fit for.
Daniel and the AInexfinder editorial team research DALL-E 3 using public product information, listing evidence, and (when available) hands-on checks. Scores reflect listing completeness and transparency — not paid placement.
krea2.org is a browser image-generation workspace that runs several third-party models side by side, including Krea 2 tiers, FLUX.2, GPT Image 2, Nano Banana and Seedream 5.0, with reference uploads, six aspect ratios and live model switching.
GenApe is an all-in-one AI platform covering writing, chat, image and video generation, image editing, product photography, translation, and research — with 145 free credits on signup and commercial rights granted to paid users.
Adobe Firefly generates images, video, and audio from models trained on licensed Adobe Stock and public domain content, with Content Credentials on every output, Firefly Boards for ideation, Custom Models for brand style, and third-party partner models alongside Adobe's own.
GPT Image 2 launched on 21 April 2026 as ChatGPT Images 2.0 and as the gpt-image-2 API model. Its distinguishing capabilities are dense, accurate text rendering, layout fidelity for infographics and packaging, up to 3840px output, prompt-guided masking, and reasoning applied to image structure before generation.
OpenArt is an all-in-one AI creation platform covering image, video, voice and audio in one place, built around a Director that lets you lay out a storyboard and shape each shot through conversation, alongside persistent characters and worlds for consistency across a project.
Stockimg AI offers separate generators for logos, stock photos, illustrations, wallpapers, product photos and social posts, then schedules the output to Instagram, TikTok, LinkedIn, YouTube and X, with an API for programmatic use.
