Skip to content
OpenAI

GPT-Image-2.6

OpenAI's image model in two API flavors, Flare for speed and Sunburst for precision edits, behind ChatGPT Images 2.6. Now with animated GIFs.

Quality

Modality

image

Resolution

4K (above 2560×1440 experimental)

Access

closed

Fabian's Take

FM

"It solidifies the lead this model series already had. It's hard to tell how much plain generation and editing improved over 2.5, but it feels faster, and it handles a long series of edits from one prompt flawlessly. It can also make animated GIFs now, which is neat, even if it's not the most important thing for business. It's still the one I reach for, for creative work as much as for the editing that goes into UI design."

OpenAI shipped GPT-Image-2.6 on 4 November 2026, two months after GPT-Image-2.5. In ChatGPT it appears as ChatGPT Images 2.6, in ChatGPT and Codex on every plan. In the API it comes as two models again, at the same token prices as 2.5.

Flare and Sunburst

gpt-image-2.6-flare is the fast one, aimed at iterating: OpenAI says it can return an image in about six seconds at low quality. gpt-image-2.6-sunburst is for detailed visual work, where the image has to stay consistent across rounds of edits. OpenAI says both generate about 50% faster than their 2.5 versions. They share one price list, so you’re choosing on speed against precision, not on budget. In ChatGPT you don’t get the choice; OpenAI doesn’t say which of the two runs there. If you want Sunburst specifically, use the API or a platform that offers it, such as Higgsfield.

What’s new

Two features you can see. ChatGPT Images 2.6 makes animated GIFs, and it lets you keep reference images of people, pets and characters and reuse them in later images. The GIFs are fun; for business work they’re a nice extra rather than a reason to switch.

The change I notice more is in long sessions. The old failure mode of conversational image editing, where the fourth edit quietly undoes the second, was already rare with 2.5. With 2.6 I can give it a whole series of edits in one prompt, and in my use it works through them flawlessly.

What it costs

Image pricing is token-based rather than per-image, which makes budgeting awkward. The published rates are $5 per million text-input tokens, $8 per million image-input tokens, and $30 per million image-output tokens, with cached input cheaper. What any single picture costs depends on its dimensions, the quality setting, and how many reference images you fed in.

Where it stands

It’s hard to say how much plain generation and editing improved over 2.5. It feels faster, and it keeps the lead the series already had. It covers two jobs that usually need different tools: creative image work, and the unglamorous editing that goes into UI design. For a second opinion I run the same request through Nano Banana. Midjourney is still the one to reach for when you want a particular look rather than a particular idea.

The Verdict

Best for: Creative image work and UI design alike: conversational editing, long runs of edits, and text-heavy visuals that have to stay legible.

Pros

  • Handles a long series of edits from one prompt without losing the thread, in my use
  • Makes animated GIFs, and keeps reference images of people, pets and characters for reuse
  • Both API models are about 50% faster than before, according to OpenAI; Flare can return an image in about six seconds at low quality
  • Same token prices as GPT-Image-2.5

Cons

  • In ChatGPT you can't choose between Flare and Sunburst
  • Still short of Midjourney on pure artistic character
  • Token-based billing makes per-image cost hard to predict

Specs

  • Pricing $5/M text input, $8/M image input, $30/M image output · cached $1.25/M text, $2/M image
  • Cost Tier moderate
  • ⚡
    Speed Tier fast
  • License Proprietary
Developer Docs