Names, versions and owners change fast. Checked October 2026.
In plain words
DALL-E was a text-to-image model from OpenAI, the company behind ChatGPT. You typed a description, such as "a cat reading a newspaper", and it drew a matching picture. The name blends the painter Salvador Dalí with WALL-E, the Pixar robot.
There were three versions: DALL-E in January 2021, DALL-E 2 in April 2022 and DALL-E 3 in late 2023, which was built into ChatGPT. Each made sharper pictures and followed instructions more closely.
In March 2025 ChatGPT switched to making images with its own GPT model, an approach OpenAI calls GPT Image. OpenAI then retired DALL-E: it was removed for developers in May 2026, and the DALL-E GPT inside ChatGPT closed in August 2026. Apart from that GPT, ChatGPT had already been making its pictures with GPT Image since March 2025.
Everyday example
A painter you describe things to
Think of telling a street painter at Marine Drive, "a monsoon evening, a red umbrella, the sea behind". They have seen thousands of evenings, umbrellas and seas, and combine them into a new painting. DALL-E did this from text alone, with pictures it learned from, in seconds.
Where the example breaks: A painter has seen the real sea. DALL-E only learned how images and captions tend to go together, so it often got hands, text and counting wrong.
Latestas of October 2026
- Status
- Retired: removed from OpenAI's API on 12 May 2026; the DALL-E GPT in ChatGPT closed on 30 August 2026
- Replacement
- GPT Image; OpenAI's API notice points developers to gpt-image-2, gpt-image-1 or gpt-image-1-mini
These details change often. The rest of this page explains what it is, which stays true.
How it actually works
The three DALL-E versions used different methods, which tracks how image AI moved in those years.
DALL-E (2021): pictures as tokens
The first version treated an image as a sequence of tokens and used a transformer to predict them after the caption, much like predicting the next word.
DALL-E 2 (2022): from noise to picture
The second version switched to a diffusion model: start with random noise, then remove it step by step, guided by an embedding of the caption.
DALL-E 3 (2023): better at following words
The third version focused on following long, detailed prompts and was placed inside ChatGPT, which rewrote your request into a fuller description before drawing.
GPT Image (2025 onward)
OpenAI's newer approach makes images with the same kind of multimodal GPT model that handles conversation, so it can use the whole chat as context and edit pictures through follow-up messages.
- Made by
- OpenAI
- Versions
- DALL-E (5 January 2021), DALL-E 2 (6 April 2022), DALL-E 3 (September 2023)
- Retired
- Removed from OpenAI's API on 12 May 2026; the official DALL-E GPT in ChatGPT closed on 30 August 2026
- Replaced by
- GPT Image models; ChatGPT has made images with GPT models since March 2025
Questions people ask
Is DALL-E still available?
No. OpenAI retired it in 2026: the DALL-E models left the API on 12 May 2026, and the official DALL-E GPT inside ChatGPT closed on 30 August 2026. ChatGPT had already switched to making images with GPT Image.
Is DALL-E the same as ChatGPT?
No. ChatGPT is OpenAI's chat app. DALL-E was a separate image model that ChatGPT used for drawing between 2023 and 2025. Image making in ChatGPT moved to GPT Image in 2025.
Where does the name DALL-E come from?
It combines the surrealist painter Salvador Dalí with WALL-E, the robot from the Pixar film. It is usually pronounced like "Dali".
What are alternatives to DALL-E?
Other text-to-image tools include Midjourney, Stable Diffusion, Google's image models in Gemini, and OpenAI's own GPT Image inside ChatGPT. They differ in style, cost and whether you can download them.