Updated
What is GPT Image 2.6?
GPT Image 2.6 is OpenAI's image generation and editing model, released November 4, 2026 and callable on UnificAlly as gpt-image-2.6-flare and gpt-image-2.6-sunburst. It is the follow-up to GPT Image 2.5 and keeps the same two-tier setup.
gpt-image-2.6-flare is the fast tier, built for everyday generation where turnaround matters. gpt-image-2.6-sunburst is the quality tier: it spends a bit more time for finer detail and follows reference images more closely. Both take a prompt and up to 16 reference images and return an image at 1K, 2K or 4K.
The main change from 2.5 is image quality. OpenAI presents 2.6 as a quality upgrade, and early leaderboard results place it at or near the top for image generation.
Key features of GPT Image 2.6
- Two tiers, one request shape. Flare for speed, Sunburst for detail. Same parameters, so switching is a one-word change in the URL.
- Better image quality. A step up from GPT Image 2.5 in overall image quality, according to OpenAI and early public rankings.
- Closer reference adherence on Sunburst. Products, faces and other details from your reference images carry into the result more faithfully.
- Up to 16 reference images. Compose several products into one shot, or edit a single photo with a plain instruction.
- 1K, 2K or 4K output. Pick the size per request, with 15 aspect ratios from 3:1 banners to 1:3 columns.
- Drop-in for 2.5. The body is identical to GPT Image 2.5, so existing integrations only need the new model name.
What GPT Image 2.6 can do
House-generated examples on both tiers.
Water and weather that hold up
A sailor hauls a rope while a wave breaks over the bow. The spray, the foam on the deck, the water running off the rails and the lightning in the distance all read as one storm, and the scene only works because the water looks right.
Signs, menus and posters spelled right
Three pieces of text in one street scene: a gold window sign, a chalkboard menu with prices and a gig poster on the wall. Every word and price comes out as written, and the text sits on real surfaces in the rain, not pasted on top.
Neon light on chrome and wet stone
Pink and teal neon bounce off a chrome tank, the exhaust and the puddles on the cobblestones, and the reflections bend with each surface. This is the kind of lighting and material work that usually needs a 3D render.
Put a product photo into a new scene
One studio photo of a red mug goes in as a reference, and the prompt moves it to a sunny cafe table with coffee and a croissant. The color and gloss of the glaze carry over and the light matches the new room. The shape is not an exact copy: the result is taller and straighter than the source, so check product proportions before you ship an edit.
Catalog-ready product shots
A leather handbag on a stone plinth with soft studio light. The grain, the stitching and the brass hardware are sharp, and the set looks like a finished catalog page on the fast tier.
What GPT Image 2.6 is best for
High-volume generation
Flare keeps turnaround short for features that make many images a day.
Product and brand shots
Sunburst sticks closer to reference photos, so the product in the picture stays the real one.
Reference-based edits
Send a photo and a plain instruction, and get back the same subject with the change applied.
Large-format output
4K output for print, hero banners, and anything that gets cropped later.
GPT Image 2.6 use cases
Turn one product photo into a set of campaign images with new backgrounds, lighting and crops. Build an in-app editor that takes instructions like "make the jacket red" and returns the edited photo. Generate marketing visuals at 4K for print or large screens. Use Flare for drafts and quick variations, then run the final picks through Sunburst for the extra detail.
GPT Image 2.6 limitations
Sunburst is slower than Flare, so it is not the best fit for anything that needs an instant result. Dense or very small text in an image can still come out wrong. Keeping one character or brand element exactly the same across separate generations is not guaranteed. 2K and 4K cost more and take longer than 1K, and each reference image adds to the price.
Flare vs Sunburst
Pick Flare by default: it is faster, and the quality is good enough for most production traffic. Switch to Sunburst for final assets, product shots and edits where the subject must match the reference closely. Both cost the same at every size ($0.039 per 1K image), so the choice is only speed against detail.
GPT Image 2.6 API examples
Send a POST to /v1/tasks with the model id. Use openai/gpt-image-2.6-flare for Flare or openai/gpt-image-2.6-sunburst for Sunburst; the input is the same.
curl -X POST https://api.unifically.com/v1/tasks -H "Content-Type: application/json" -H "Authorization: Bearer YOUR_API_KEY" -d '{
"model": "openai/gpt-image-2.6-flare",
"input": {
"prompt": "A ceramic coffee mug on a sunlit kitchen counter, soft morning light",
"aspect_ratio": "4:3",
"resolution": "2K"
}
}'
To edit or compose from photos, add image_urls with up to 16 image links. The prompt can be up to 5,000 characters. The request returns a task_id right away; poll the task status with it until the image is ready. Every parameter is listed in the GPT Image 2.6 API docs.
FAQs
People also ask about GPT Image 2.6
GPT Image 2.6 is OpenAI's image generation and editing model, released November 4, 2026. It follows GPT Image 2.5 and comes as two models, gpt-image-2.6-flare and gpt-image-2.6-sunburst. Both take a text prompt, with or without reference images, and return an image.
Yes. Both gpt-image-2.6-flare and gpt-image-2.6-sunburst are callable on UnificAlly through the REST API and the playground on this page.
Flare is the fast tier and the right default for most work, like high-volume generation and quick drafts. Sunburst is the quality tier. It takes a little longer and gives finer detail, and it sticks closer to your reference images, which helps for product shots and edits where the subject has to stay the same.
OpenAI describes 2.6 as a quality step up over 2.5, and early public leaderboard results put it at or near the top for image generation. The API shape is the same as 2.5, so moving over is a change of model name.
Flare and Sunburst cost the same: $0.039 per 1K image. 2K and 4K cost more, and each reference image adds a small charge. The full table is on this page and on /pricing.
1K, 2K and 4K, with 1K as the default. There are 15 aspect ratios, from 1:1, 4:3, 3:2, 16:9 and their portrait versions to wide 21:9, 2:1 and 3:1 and tall 9:21, 1:2 and 1:3.
Up to 16 per request, through image_urls. With references the model edits or composes from them; without, it generates from the prompt alone.