What can GPT Image 2.6 do?

It makes a still image, a short GIF, or an SVG, and a later edit can keep the same character, product, or brand details.

GPT Image 2.6 is the image model OpenAI announced on 4 November 2026. It is about 50% faster than the previous image model. In ChatGPT, in Codex, and through the API, you can ask for a picture, a loop, or a flat vector file. This page is an independent reading of that announcement, plus a box that calls the API when a key is set.

A faster first picture

Flare is the faster of the two models. At low quality it can return a picture in about six seconds, and it is stronger on posters and product layouts. Use it when you want to see whether the idea works. Sunburst is the more detailed model. At medium or high quality, a later edit stays closer to the character, product, or brand you already have.

A GIF, or an SVG

A GIF here is 9 frames at 200 milliseconds. Ask for a mascot wave, a reaction, or a product that turns and comes back to the first frame. An SVG here is one file. Ask for flat shapes, a wide margin, and no texture, so it can sit on a site or a printed page.

Those two formats are generated on their own. They do not go into the markup step that edits one part of a photo.

An edit of one spot

Markup points at the part you want changed and leaves the rest alone. The mark is a transparent PNG the same size as the picture. It applies to the first image only, and it cannot be combined with a mask. For most edits, start at low complexity. Raise complexity only when the scene has many parts. Quality is still the control that makes the picture sharp.

OpenAI reports 4 billion images a week across ChatGPT, Codex, and the API. That is their number, not a count from this site. The announcement itself is Introducing Images 2.6.