DALL-E (Agent)
Last reviewed
Jun 7, 2026
Sources
No citations yet
Review status
Needs citations
Revision
v4 · 1,160 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 7, 2026
Sources
No citations yet
Review status
Needs citations
Revision
v4 · 1,160 words
Add missing citations, update stale details, or suggest a clearer explanation.
DALL-E (Agent) refers to the official DALL·E GPT published by OpenAI inside ChatGPT, a first-party agent variant of ChatGPT that exposes DALL·E 3 image generation through a dedicated conversational interface. The agent is reachable at the URL chatgpt.com/g/g-2fkFE8rbu-dall-e and is operated by OpenAI rather than a third-party developer.[1] It is distinct from DALL·E itself, the underlying family of text-to-image diffusion and transformer models, which is covered at dall e. The DALL·E GPT became a separate, persistent endpoint after OpenAI replaced DALL·E 3 as ChatGPT's default image generator with GPT Image (the GPT-4o-native image stack) in March 2025, leaving the dedicated GPT as the primary way for ChatGPT users to reach the older DALL·E 3 model from the consumer product.[2]
This page disambiguates the agent or wrapper from the model. The DALL·E model (text-to-image generation, originally a 12-billion-parameter GPT-3-derived system, later succeeded by DALL·E 2 and DALL·E 3) is documented at dall e and the third-generation variant at dall e 3. The DALL·E GPT, by contrast, is a productized chat agent that lives in the ChatGPT Store (see also gpt store) and that uses the conversational tool-use loop introduced with ChatGPT's image-generation tool in late 2023 to route prompts to the underlying image model.[3][4]
The distinction matters because the agent and the model have separate launch dates, separate update cadences, and separate user experiences. The model launched in three generations between January 2021 (DALL·E 1) and late 2023 (DALL·E 3).[5] The agent, in the sense of a first-class GPT object selectable from the ChatGPT GPT picker, only became prominent after the GPT Store opened on January 10, 2024, and grew in importance after March 2025, when DALL·E 3 was no longer the default image generator inside the main ChatGPT chat.[2][6]
OpenAI announced DALL·E 3 on September 20, 2023, and made it generally available inside ChatGPT Plus and ChatGPT Enterprise in early October 2023.[7][8] In that initial integration, DALL·E 3 was not a separate GPT; it was a tool callable from the default ChatGPT conversation, with the language model acting as a prompt engineer that turned the user's request into a longer, more descriptive caption before sending it to the image model.[4] OpenAI described this as part of an "ask-as-you-would-a-person" workflow in which the user does not need to write image-style prompts directly.[4]
When OpenAI launched the GPT Store on January 10, 2024, GPTs (called custom GPTs at the time, see custom gpt and custom gpts) became browseable, shareable, and rankable agents.[6] OpenAI itself maintained a small number of first-party GPTs, of which the DALL·E GPT is one. The DALL·E GPT carries a builder attribution of "By ChatGPT" or "By OpenAI" and uses the same image-generation tool that the broader ChatGPT product calls, but it is configured with a system prompt that focuses the conversation on image creation and revision rather than open-ended chat.[1][9]
The DALL·E GPT became materially more visible in March 2025, when OpenAI rolled out GPT Image (the GPT-4o native image generator) as the default image-generation path inside ChatGPT for Plus, Pro, Team, and Free users.[2] DALL·E 3 was no longer reachable by default from ordinary chats, but it remained available to ChatGPT users who explicitly opened the DALL·E GPT, and it remained available through the OpenAI API.[2] In effect, the DALL·E GPT became the canonical consumer surface for the DALL·E 3 model after that switchover, while AI image generation in the default ChatGPT experience was handled by GPT Image.
The DALL·E GPT is a thin agent over DALL·E 3. Its capabilities are therefore the capabilities of the underlying model, plus the prompting and tool-use behaviour OpenAI has wired into its system prompt. Users open the GPT from the ChatGPT app, the web sidebar, or by direct URL, then describe an image in natural language. The agent rewrites the request into a longer prompt of roughly 100 words, calls the image tool, and returns one or two generated images per turn.[9] DALL·E 3 itself was trained to follow long prompts closely, to render text inside images (a known weakness of DALL·E 2), and to produce coherent compositions with multiple subjects.[7]
Reported user-facing features include:
The DALL·E GPT does not expose finer-grained model parameters (sampler choice, guidance scale, seed) the way the OpenAI API does; those controls remain in the API-level images/generations endpoint.[11]