DALL-E: What It Was, Why OpenAI Retired It & What to Use Now
If you searched for DALL-E recently, you may have noticed something odd: the tool that put AI image generation on the map is gone. DALL-E, OpenAI’s pioneering text-to-image model, was removed from the OpenAI API in May 2026, and the official DALL-E GPT inside ChatGPT closed in August 2026. If you arrived here looking for a download link, this guide will save you a wasted afternoon — and point you at what actually works now.
The good news is that nothing you learned from DALL-E is wasted. Its successor, built into ChatGPT, works the same conversational way and does more. This guide explains what DALL-E was, how it worked, why OpenAI replaced it, and what to use instead.
What is DALL-E?
DALL-E is OpenAI’s family of text-to-image models — software that turns a written description into a picture. The name is a blend of the surrealist artist Salvador Dalí and Pixar’s robot WALL-E, and it fits: the model was famous for dreamlike, imaginative images from plain-language prompts.
There were three generations. The original DALL-E arrived in January 2021, DALL-E 2 followed in April 2022 with much sharper images, and DALL-E 3 launched in September 2023. It was DALL-E 3 that most people remember, because it was the first version built directly into ChatGPT.
That integration was the breakthrough. Instead of learning a special prompt language, you simply described the image you wanted in a chat message — “a cosy bookshop at dusk in the rain, warm light in the windows” — and ChatGPT rewrote your request into a detailed prompt before DALL-E 3 drew it. For millions of people, it was their first experience of making images with AI.
How DALL-E 3 worked

Using DALL-E 3 was deliberately simple — no software to install, no separate app to learn:
- Open ChatGPT and describe the image you want in plain words.
- ChatGPT expands your description into a detailed prompt automatically.
- DALL-E 3 generates the image, usually in under a minute.
- If the result is not right, ask for a change in a follow-up message — “make the background darker” or “change the dog to a cat” — instead of starting over.
Images came in three fixed sizes — square 1024 by 1024, or tall and wide 1024 by 1792 — in standard or HD quality. Free ChatGPT users could generate a small number of images a day; from August 2024 the published allowance was two images a day. Plus subscribers, paying $20 a month, got much higher limits. Developers could also call DALL-E 3 through the OpenAI API, paying per image — $0.04 for a standard square image and up to $0.12 for an HD tall or wide one.
Its biggest strength was prompt accuracy: DALL-E 3 followed a brief more literally than its rivals, and it rendered short text inside images more reliably than earlier models. That made it the go-to choice for blog illustrations, social graphics and product mockups — the jobs where the image has to match the words.
Why OpenAI retired DALL-E
DALL-E was not retired because it failed. It was retired because OpenAI built something better, and keeping both made no sense. The timeline tells the story:
- March 2025: OpenAI launches GPT Image 1 (gpt-image-1), a natively multimodal model that generates and edits images inside ChatGPT, with no separate prompt-rewriting step.
- December 2025: ChatGPT users are migrated automatically to GPT Image 1.5.
- 4 March 2026: DALL-E 3 is retired on Microsoft’s Azure OpenAI service.
- 12 May 2026: DALL-E 2 and DALL-E 3 are removed from the OpenAI API. Developers were notified in November 2025, and any code still calling the old model names now returns an error.
- 30 August 2026: the official DALL-E GPT inside ChatGPT is retired, and users are pointed to ChatGPT Images instead.
The technical reason is straightforward. DALL-E 3 was a separate image model bolted onto a chat model: ChatGPT wrote a prompt, then handed it over. The GPT Image models are one model that understands text and images together, so they edit existing pictures, keep a character consistent across images, and render text far more accurately.
What replaced DALL-E: ChatGPT Images and GPT Image 2.5

For everyday users, the replacement is simply ChatGPT Images — the image feature inside ChatGPT. If you used DALL-E 3 inside ChatGPT, you are already using its successor: describe an image, or upload a photo and ask for an edit, in the same chat window. ChatGPT Images 2.0 launched in April 2026, and version 2.5 followed in September 2026.
Behind the scenes, the current models are GPT Image 2.5 Flare, built for fast everyday generation, and GPT Image 2.5 Sunburst, built for work where editing precision matters most. Unlike DALL-E 3’s flat per-image price, these models are billed by tokens: at the time of writing, $5 per million text input tokens, $8 per million image input tokens and $30 per million image output tokens. For most people that detail does not matter — you just use ChatGPT — but it matters if you are building an app.
Microsoft’s Bing Image Creator, which also offered DALL-E 3, now offers GPT-4o image generation as well, so the free route through Microsoft still exists in a new form. And Google’s Gemini image model — the one behind the “Nano Banana” nickname — is a strong alternative for editing real photos; our Gemini Nano Banana guide explains how it works.
How to generate images in ChatGPT now
The workflow you knew from DALL-E 3 still applies, with one difference: you can now start from an image as easily as from words.
- Open ChatGPT at chatgpt.com or in the ChatGPT app.
- Describe the image you want, or upload a photo and describe the change.
- Refine it with follow-up messages — one change at a time gives cleaner results.
- Download the image you like. Free accounts get a small daily allowance; Plus, at $20 a month, raises the limits considerably. Our guide to whether ChatGPT Plus is worth it will help you decide before you pay.
One practical note: the free tier’s image allowance is shared with other features and changes from time to time, so if image generation suddenly stops working for the day, you have most likely hit the daily cap rather than a bug. Wait until the next day, or switch to a free alternative such as Microsoft’s image creator or one of the tools in our free AI photo editor guide.
Prompt tips that still work

Everything that made a good DALL-E prompt still makes a good ChatGPT Images prompt:
- Be specific about three things: the subject, the setting and the style. “A golden retriever puppy in a sunny meadow at golden hour, photography style” beats “a dog” every time.
- Put text in quotes. If you want words in the image, write them in quotation marks — “a sign that reads OPEN” — and keep them short.
- Change one thing at a time. Ask for the background change first, then the lighting, then the colours. Bundling five changes into one message muddles the result.
- Name the style precisely. “Studio product photo, soft daylight, plain grey background” gives the model far more to work with than “make it look professional”.
- Generate two or three variations. Image models are inconsistent by nature; keeping the best of three is normal practice.
DALL-E vs the alternatives today
If you are choosing an image tool now, the landscape has settled into a few clear options. ChatGPT Images is the natural home for anyone who already uses ChatGPT — it is the direct successor, it edits photos as well as generating them, and it needs no new account. Google’s Gemini image model is the strongest rival for editing real photos, especially keeping a person’s face consistent across edits. Midjourney remains the favourite for artistic, painterly styles, though it runs through Discord and has a steeper learning curve.
If you still have DALL-E code
For developers, the migration is a small job with one gotcha. Replace the dall-e-3 model name with gpt-image-2.5-flare for fast, high-volume work or gpt-image-2.5-sunburst for precise edits. The bigger change is in the response: DALL-E calls returned a web link to the image, while the GPT Image models return the picture itself as base64 data, which your code must decode and save.
Frequently asked questions
Is DALL-E still available?
No. DALL-E 2 and DALL-E 3 were removed from the OpenAI API on 12 May 2026, and the official DALL-E GPT inside ChatGPT was retired on 30 August 2026. Use ChatGPT Images instead — it is the same chat workflow, powered by the newer GPT Image models.
Was DALL-E free?
Yes, in limited amounts. Free ChatGPT users could generate a small number of images a day with DALL-E 3, and Plus subscribers at $20 a month got much higher limits. ChatGPT Images follows the same pattern.
What replaced DALL-E?
OpenAI’s GPT Image family replaced it. In ChatGPT that means ChatGPT Images, currently version 2.5, with the GPT Image 2.5 Flare and Sunburst models behind it. Microsoft’s Bing Image Creator now offers GPT-4o image generation as its free route.
Can I still use the images I made with DALL-E?
Yes. Under OpenAI’s terms, you own the images you generated with DALL-E, so anything you downloaded remains yours to use. Only the ability to generate new images with the old models has gone.
Is ChatGPT Images better than DALL-E 3?
For most jobs, yes. It renders text in images more accurately, edits existing photos, keeps characters consistent across images, and understands follow-up instructions in the same conversation. DALL-E 3’s strength was prompt accuracy for its time — and its successor kept that while adding far more.
Final word
DALL-E mattered because it made image generation ordinary: type a sentence, get a picture. It taught millions of people to describe images in words, and that skill is exactly what its successor rewards. Do not waste time hunting for a DALL-E download or an old API key — both are gone. Open ChatGPT, describe the image you want, and refine it in conversation, just as you did before. If you want to compare the wider field, our free AI photo editor guide and our Gemini Nano Banana guide are good next reads.
Sources: OpenAI — https://openai.com/



