GPT Image 2.6 團隊The short answer: Yes — OpenAI released GPT Image 2.6 on November 4, 2026, and both models are live now. You will see two model IDs, not one: gpt-image-2.6-flare for fast work and gpt-image-2.6-sunburst for detail and precise edits, with no plain gpt-image-2.6 anywhere. You can reach it three ways: ChatGPT, the Image API through generations and edits, or a browser generator that needs no API key. Most pages that still say "not out" were written before November 4.
The changelog says it plainly
The changelog entry is one sentence long, and it settles the question by itself.
Released GPT-Image-2.6 Sunburst and GPT-Image-2.6 Flare for image generation and editing, with adjustable complexity, SVG generation, and animated GIF generation. Added markup-guided editing for supported GPT Image 2.5 and 2.6 models.
The line sits under Nov 4 in the November 2026 block. It tags two endpoints as well, v1/images/generations and v1/images/edits. One clause in that quote is easy to misread: markup editing also works on 2.5. It shipped alongside 2.6, but it is not a 2.6 exclusive.
Two model IDs, and no plain gpt-image-2.6
OpenAI split the release into two variants.
gpt-image-2.6-flare |
gpt-image-2.6-sunburst |
|
|---|---|---|
| Role | Smaller model | Base model |
| Built for | Fast iteration | Detail and precise edits |
| Image generation | Yes | Yes |
| Image editing | Yes | Yes |
| Transparent background | Yes | Yes |
A request has to name one of the two IDs, because gpt-image-2.6 on its own does not exist. OpenAI's migration table draws the line by your current setup: 2.5 Flare moves to 2.6 Flare, and 2.5 Sunburst moves to 2.6 Sunburst. A workflow that already cleared its quality bar on GPT Image 2 starts on Flare, and one that fell short on a complex task starts on Sunburst.
What the API accepts
Both variants read the same request fields.
| Field | Accepted values |
|---|---|
quality |
auto, low, medium, high, xhigh, max |
complexity |
low, medium, high (defaults to high) |
size |
auto, 1024x1024, 1536x1024, 1024x1536 |
background |
auto, opaque, transparent |
output_format |
png, jpeg, webp; svg and gif on generations only |
markup |
Transparent PNG overlay on edits, same pixel size as the first input |
complexity decides how much the model plans before it renders, and quality controls rendering detail such as texture and small lettering. A custom size needs both dimensions as multiples of 16, with an aspect ratio between 1:3 and 3:1.
GIF output adds two fields. Set frame_count somewhere between 2 and 64, and frame_duration_ms between 20 and 2000. Leave both out and you get 9 frames at 200ms each. The edit endpoint takes neither svg nor gif, so a looped animation has to come from a generation call.
Where to use it
There are three practical doors, and only one of them asks for an API key.
| Surface | What you get | API key |
|---|---|---|
| ChatGPT | Available in the product | No |
| Image API | generations and edits |
Yes |
| This site | 1K, 2K and 4K stills plus GIFs | No |
The browser generator runs both variants and accepts up to 16 reference images. Quality is low, medium or high, and high adds 50% to the credit price. A failed generation is refunded on its own.
| Output | Flare credits | Sunburst credits |
|---|---|---|
| 1K still | 20 | 26 |
| 2K still | 30 | 39 |
| 4K still | 60 | 78 |
| Animated GIF | 40 | 52 |
Those are the low and medium rates, so high costs half again as much. Packs are listed on the pricing page, and the API reference lists both model IDs and their token rates.
On the OpenAI side, standard rates are $8 per million image input tokens, $30 per million image output tokens and $5 per million text input tokens. GPT Image 2.5 charges exactly the same. Equal rates do not mean an equal bill, because the two models can burn different numbers of tokens. The 2.6 vs 2.5 comparison walks through that gap.
Same name, different company: MAI-Image-2.6
Microsoft AI publishes a model called MAI-Image-2.6, and it has nothing to do with this release. It belongs to Microsoft, not OpenAI, and it has its own documentation. Same version number, different vendor. If a tutorial says MAI-Image-2.6, it is not describing GPT Image 2.6, however the headline reads.
Why the search results still say no
Most pages ranking for this question were written for older releases, and their dates give them away. The heavy hitters cover the GPT Image 2 launch in April 2026, or the 2.5 launch in September 2026. Both predate November 4.
That is the whole reason the question still gets asked. A page about GPT Image 2 cannot tell you about 2.6, and a page last updated in September stops at 2.5. This one carries a revision log for the same reason: the answer will move again.
FAQ
Is GPT Image 2.6 out in ChatGPT?
Yes. ChatGPT is one of the ways to reach it; the model IDs in this post are what you send in code.
Can I try GPT Image 2.6 without an API key?
Yes. The generator on this site runs both variants in a browser. You do not need an OpenAI account, a key, or a paid API tier to make a first image.
Does GPT Image 2.6 have a Batch rate?
No. OpenAI's Batch price table lists GPT Image 2, both 2.5 variants and older models, and no 2.6 model appears in it. The 2.6 model pages list no Batch price. A pipeline that leaves 2.5 Batch behind pays the standard $30 per million output tokens instead of $15.
Why do some pages still say GPT Image 2.6 is not out?
They were published before November 4, 2026, and many have not been touched since. Check the date at the top of a page, then compare it with the changelog date.
Which model ID should I start with?
Start with gpt-image-2.6-flare. It is the smaller variant, built for fast iteration, and the migration table routes GPT Image 2 workflows that already cleared their quality bar to it. Move to Sunburst when detail or edit accuracy matters more.
Sources
- API changelog — the November 4, 2026 release note, quoted in full above.
- Image prompting guide — the 2.6 parameter table, the complexity default, custom size rules and the migration table.
- Image generation guide — SVG and GIF output, the GIF frame fields, markup edits on 2.5 and 2.6, and the cost-per-image caveat.
- gpt-image-2.6-flare and gpt-image-2.6-sunburst — model IDs, supported endpoints and the missing Batch price.
- gpt-image-2.5-sunburst — the $15 Batch rate used for comparison.
- API pricing — standard image and text token rates.
- MAI-Image-2.6 — Microsoft AI's model with the same version number.
Revision log
- November 5, 2026 — Page published.