ChatGPT gets native GPT-4o image generation

Background

Before March 2025, making pictures inside ChatGPT was a job for DALL·E. DALL·E 3 had been ChatGPT's image model since October 2023 ^[3]. In that arrangement the chat and the drawing were handled by two different models: one held the conversation, and when the user asked for an image, the product reached for a dedicated image model. That is why image making felt like a separate tool attached to the side of the chat window rather than part of the conversation itself.

What happened

On 25 March 2025, OpenAI released native image generation in GPT-4o for ChatGPT ^[1]. Within that month, image generation in ChatGPT moved from DALL·E to GPT Image ^[2][3]. DALL·E 3, which had held the role since October 2023, thereby became the legacy option ^[3].

"Native" here means the image capability sits inside the same model that carries on the conversation, instead of being called in from a separate image model. The switch was also a change of name and of lineage: the feature was no longer described as a DALL·E feature inside ChatGPT, but as GPT Image.

Details at launch

Access was rolled out in stages. Plus, Pro, Team and Free users got the feature first, with OpenAI saying that Enterprise and Edu access would follow soon after ^[4]. DALL·E did not disappear at launch: it remained reachable through a dedicated DALL·E GPT ^[4].

The trade-off was speed. Because the model produced more detailed pictures, images often took up to a minute to render ^[4]. A minute is a long pause in a chat window, and it says something about where the cost of this kind of generation sits: the detail in the output is bought with waiting time.

Reaching developers

On 23 April 2025, OpenAI brought the same model to developers as gpt-image-1 ^[5]. Until then the new image capability was something you met inside ChatGPT; with the API release, the same model became something a program could call. Because it was the same model rather than a parallel one, what users saw in the chat product and what developers could build against were meant to be the same thing.

Afterwards

The line continued. OpenAI later shipped gpt-image-1.5 in December 2025 and gpt-image-2 in April 2026 ^[1]. The naming stayed with the gpt-image family, so the models arriving later read as successors to the one introduced in April rather than as new DALL·E releases.

Why it matters

The event marks a change in how OpenAI organised image generation. For roughly a year and a half, DALL·E had been the name attached to image making in ChatGPT; from March 2025 the capability belonged to the main GPT model line, and DALL·E was kept as a legacy path ^[3][4]. Second, the gap between the consumer launch and the API launch was short: the model appeared in ChatGPT on 25 March 2025 and reached developers as gpt-image-1 on 23 April 2025 ^[1][5]. A reader tracking the field can use those two dates to see how a product feature and a developer-facing model now tend to arrive close together, and can follow the later version numbers — 1.5 and 2 — as the continuation of a single named series ^[1].