Two models, one page
Flare is the fast one for drafts and volume. Sunburst is slower and built for detail and for edits that should leave the rest of the image alone. There is no single gpt-image-2.6 ID.
Try GPT-Image 2.6 online or via API. OpenAI's newest image models: Flare for fast drafts, Sunburst for edits that keep products and characters intact. Text-to-image and reference editing from 1K to 4K.
| Specification | Credits | Price (USD) |
|---|---|---|
| Model & ModalityGPT-Image 2.6 FlareOutput FormatGIFResolution1K | 2per generation | $0.02per generation |
| Model & ModalityGPT-Image 2.6 FlareOutput FormatPNGResolution1K | 2per generation | $0.02per generation |
| Model & ModalityGPT-Image 2.6 FlareOutput FormatPNGResolution2K | 2.5per generation | $0.025per generation |
| Model & ModalityGPT-Image 2.6 FlareOutput FormatPNGResolution4K | 3.5per generation | $0.035per generation |
| Model & ModalityGPT-Image 2.6 SunburstOutput FormatGIFResolution1K | 2per generation | $0.02per generation |
| Model & ModalityGPT-Image 2.6 SunburstOutput FormatPNGResolution1K | 2per generation | $0.02per generation |
| Model & ModalityGPT-Image 2.6 SunburstOutput FormatPNGResolution2K | 2.5per generation | $0.025per generation |
| Model & ModalityGPT-Image 2.6 SunburstOutput FormatPNGResolution4K | 3.5per generation | $0.035per generation |
Prices follow the model configuration. Reference images are not billed unless the model says so.
GPT-Image 2.6 is OpenAI's image model family released on November 4, 2026. Like 2.5 it ships as two models, Flare and Sunburst, and both are on this page.
Flare is the fast one for drafts and volume. Sunburst is slower and built for detail and for edits that should leave the rest of the image alone. There is no single gpt-image-2.6 ID.
Write a prompt, or add up to 10 reference images to edit a photo or combine several into one.
Pick 1K, 2K, or 4K and auto or one of 13 fixed aspect ratios. Each request returns one image.
Switch Output to GIF for a short looping animation: 2 to 9 frames, 100 to 500 ms per frame, in auto or 7 aspect ratios. GIFs are made from the prompt alone, without reference images.
Same two-model split, same endpoints. The changes are speed on Flare, planning control, and new output types on OpenAI's API.
OpenAI says Images 2.6 generates about 50% faster, and that Flare can return a low-quality image in as little as six seconds. In an independent 80-image test on launch night, 2.6 Flare finished every prompt about twice as fast as 2.5 Flare at the same size and quality.
OpenAI positions Sunburst for keeping details consistent from one edit to the next at medium or high quality. The same independent test found 2.6 Sunburst no faster than 2.5 Sunburst, so pick it for control, not for wait time.
2.6 plans the image before rendering. OpenAI's default is the highest planning level; lower levels suit focused edits, higher levels suit posters, infographics, and scenes with many objects.
2.6 adds animated GIF output, and you can use it on this page today. OpenAI's Image API also returns SVG vectors and accepts a drawn markup layer to point at the area to change; those two are not available in this playground yet.
Both IDs sit in the Model Type control at the top of this page. Compare them with the same prompt and references.
Social crops, product-page drafts, posters you will iterate on, anything you generate many times before keeping one. ID: gpt_image_2_6_flare.
Brand stills, pack shots, character art that goes through several edits. Figma Weave's creative director described 2.6 as keeping characters, products, and brand details intact while generating and editing. ID: gpt_image_2_6_sunburst.
Keep the prompt, references, and size the same. If Flare is good enough, stay on it. Move to Sunburst only when the edit still drifts.
Use cases from OpenAI's launch and early API tests.
Put the headline in quotes and describe the layout. Higher planning helps when a page has many text blocks; check spelling in the result.
Upload the photo, name the one thing to change (color, label language, background), and list what must not move. Start on Sunburst.
Generate the character once, then keep editing pose, outfit, or scene with the same reference so the face and colors hold.
Pick GIF, describe the motion and the loop, and get a short animation such as a mascot waving or a product turning. Keep the motion simple: 9 frames at 200 ms is under two seconds.
One asynchronous contract for both models.
gpt_image_2_6_flare or gpt_image_2_6_sunburst.
POST /api/openapi/submit with model, a unique out_task_id, and params: prompt, optional image_url (up to 10), resolution, aspect_ratio, or size as WIDTHxHEIGHT. For a GIF, add output_format gif with frame_count (2 to 9) and frame_duration_ms (100 to 500).
POST /api/openapi/query with the same out_task_id until the status is completed or failed. Completed responses include the image URL.
Compare GPT-Image 2.6 with other image models on AIReiter.
The previous Flare and Sunburst pair. Stay here until a same-prompt test on 2.6 wins.
02GPT-Image 2OpenAI's single-model line with 1K to 4K output and 14 ratio options.
03Nano Banana ProGoogle's image model for editing and multi-image composition up to 4K.
04Seedream 5.0 ProByteDance stills with large reference sets.
| Control | Type | Default | Notes | |
|---|---|---|---|---|
| 01 | prompt | text | — | — |
| 02 | output_format | string | png | — |
| 03 | image_url | image | — | — |
| 04 | resolution | string | 1K | — |
| 05 | aspect_ratio | string | auto | — |
| 06 | frame_count | integer2 – 9 | 9 | — |
| 07 | frame_duration_ms | number100 – 500 | 200 | — |
Flare for drafts and volume, Sunburst for detailed edits. Choose GIF for a looping animation.
Up to 10 reference images. Say what changes and what stays.
For images, 1K, 2K, or 4K and auto or a fixed ratio. For GIFs, the frame count, frame duration, and one of the GIF ratios.
Run the same inputs on the other model if the result is not right.
gpt_image_2_6_flare, gpt_image_2_6_sunburst
POST to /api/openapi/submit with the model, a unique out_task_id, and parameters.
POST to /api/openapi/query with the same out_task_id until the task completes.
Save outputs in your own storage rather than hot-linking them.
OpenAI's image model family released on November 4, 2026, offered as two API models: Flare for speed and Sunburst for detail and consistent edits.
Flare for drafts and volume. Sunburst when an edit must leave the rest of the image alone or fine detail matters. There is no plain gpt-image-2.6 ID.
Flare is. OpenAI reports about 50% faster generation, and an independent launch-night test measured 2.6 Flare about twice as fast as 2.5 Flare. Sunburst was not faster in that test.
Run the same prompt and references on both first. If 2.6 Flare matches the quality, you get shorter waits. GPT-Image 2.5 stays available at /image/gpt-image-2-5.
Yes. Set Output to GIF, or send output_format gif through the API. You get 2 to 9 frames at 100 to 500 ms each, in auto, 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, or 2:3. GIFs are made from the prompt only; reference images are not supported for GIF.
Not yet. OpenAI's Image API supports SVG for 2.6, but this playground and our API return PNG images and GIFs for now.
Up to 10 per request. Use one to edit a photo, or several to combine products, characters, or styles into one image.
gpt_image_2_6_flare or gpt_image_2_6_sunburst with POST /api/openapi/submit, then poll POST /api/openapi/query with the same out_task_id.