The short answer
● Same release — both are Images 2.6, launched 4 November
● Flare — speed, drafts, as little as six seconds at low quality
● Sunburst — quality and consistency across repeated edits
● Same price — $30 per million image output tokens
● Pick by — whether you iterate or finish
OpenAI shipped two image models on the same day rather than one, which is the clearest signal yet that image generation has split into two different jobs.
|
GPT-Image-2.6 Flare |
GPT-Image-2.6 Sunburst |
| Tuned for |
Speed |
Quality and edit consistency |
| Best at |
Quick iteration and drafts |
Medium and high quality output |
| Fastest reported |
Six seconds at low quality |
Not published |
| Edit consistency |
Not its strength |
The reason it exists |
| Price |
$30 per million output tokens |
$30 per million output tokens |
| GIF and SVG |
Yes, same Images 2.6 features |
Yes, same Images 2.6 features |
How to choose
If you are exploring and will throw most outputs away, Flare. The six-second figure is a best case at low quality, but the gap is real and iteration speed compounds.
If you are producing a final asset, or making several edits to the same image and need the subject to stay recognisable between them, Sunburst. Consistency across repeated edits is the specific thing it is tuned for, and it is the thing Flare is not.
Since the published rate is the same per million output tokens, the cost difference comes from output size and quality rather than the model you pick. Choosing Sunburst for a draft costs you time; choosing Flare for a final asset costs you a redo.
What applies to both
Both carry the Images 2.6 feature set: 50% faster generation than the previous version, GIF creation, and SVG export. SVG requires n=1 and supports neither streaming, partial images, output compression nor the prompt rewriter, and the image-edit endpoint will not return it at all. That restriction is the same whichever model you call.
Sources: Hugging Face · OpenAI.
Published 8 November 2026. Every figure above was checked against the reporting linked here on the same day.