On November 4, OpenAI released Images 2.6, the latest version of the image model behind ChatGPT. For designers, this is the most practical image-model update of the year — not because of a dramatic leap in quality, but because it finally outputs two formats designers actually deliver to clients: animated GIFs and vector SVG files.
Until now, AI image tools were great at making pictures and bad at handing you a usable file. You got a flat raster image, and if you needed a looping animation or a scalable logo, you rebuilt it by hand. Version 2.6 starts to close that gap.
Two models instead of one
The API release ships as two models with the same price: GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst.
Flare is the fast one. OpenAI says it can return an image in as little as six seconds at low quality, which makes it the right pick for drafts, quick variations and brainstorming. Sunburst is the careful one: at medium or high quality, OpenAI positions it for detailed work and, more importantly, for edits that keep a character, product or brand element consistent from one version to the next.
Both cost the same per token, so the choice is about workflow, not budget. Use Flare while you explore, switch to Sunburst once the direction is locked.
OpenAI also claims images generate about 50% faster than before, without publishing the baseline behind that number. Treat it as a direction, not a guarantee.
Animated GIFs, finally native
The headline feature for social and motion designers is native GIF output. In ChatGPT, you can now ask for an animated GIF in a single step instead of generating stills and stitching them together elsewhere. Through the API, you set the output format to GIF and control the frame count (2 to 16 frames) and the timing of each frame.
That covers a lot of everyday design work: looping social posts, loading animations, product micro-interactions, animated stickers and simple cinemagraph-style loops. The results are short loops, not full video — think seconds, not scenes — but for social formats that is usually enough.
One caveat: GIF creation in ChatGPT is still rolling out, and some GIF access through the API is preview-gated, so the feature may not have reached every account yet.
SVG export: the one that matters most
SVG output is the bigger deal for brand and graphic designers. You can now ask the model for a vector file — a logo mark, an icon set, a simple illustration — that scales cleanly from a favicon to a billboard, with no trace-and-clean-up pass in Illustrator.
There are limits worth knowing. SVG works on the generation endpoint only: the edit endpoint cannot output SVG, and an SVG cannot be used as an edit input. The file is produced in two steps — the model first creates a reference image, then a second authoring pass turns it into the vector document — so expect a higher cost per file than a plain raster image. Streaming and partial images are not supported for SVG either.
In practice, treat the SVG as a strong starting point, not a finished asset. Generate the mark, then refine the paths, spacing and curves in Illustrator or Figma, the way you would with any first draft.
Two quieter upgrades
Two smaller additions deserve attention. Markup-guided edits let you draw an annotation on an image — circle the area, scribble a note — to tell the model exactly what to change, instead of describing the region in words. And a new complexity setting controls how much planning the model does before rendering, separately from image quality.
Neither is flashy, but both address the same frustration: getting an AI image to change one thing without changing everything else.
What to watch before you switch
A few honest notes before you rebuild a workflow around 2.6. The speed claims are OpenAI’s own, with no independent benchmark published. The consistency promise of Sunburst is real on paper, but test it with your own brand assets before promising a client a matching set of twenty images. And remember that pricing is per token, not per image: $5 per million text input tokens, $8 per million image input tokens and $30 per million image output tokens — the same rates as version 2.5, with SVG’s two-step process adding to the bill.
The bottom line
Images 2.6 does not replace a designer. It removes two tedious steps — turning stills into loops, and tracing rasters into vectors — that used to sit between an AI draft and a deliverable. If you make social content, try the GIF output in ChatGPT. If you work on logos and icons, test the SVG export through the API. Keep Flare for drafts and Sunburst for finals, and keep your own judgement on everything in between.
That is where these tools are heading: less time rebuilding files, more time deciding what the work should say.



