GPT-Image-2.5 Flare vs Sunburst: which one to use
OpenAI shipped ChatGPT Images 2.5 as two API models that cost the same per token. What actually separates Flare from Sunburst, and how to pick one.
Last updated: 2026-09-09
The short answer
Use Flare unless you have a reason not to. OpenAI names it the default for most applications: higher-quality images than GPT Image 2 at 50% lower latency.
Use Sunburst when an image goes through several rounds of edits and each round has to hold everything else steady. It is built for premium visual work and trades generation time for that control.
Both models bill at exactly the same token rates. Neither is the "cheap one".
What OpenAI says about each
Both arrived on 8 September 2026, the same day ChatGPT Images 2.5 rolled out to ChatGPT users.
GPT-Image-2.5 Flare
Model id gpt-image-2.5-flare. OpenAI calls it "the default choice for most applications, delivering higher-quality images than GPT Image 2 at 50% lower latency." The listed uses are creator and social content, product experiences, visual search, rapid image prototyping, and high-volume generation.
Read that last one carefully. High-volume generation is a latency problem before it is a quality problem, and Flare is the model built for it.
GPT-Image-2.5 Sunburst
Model id gpt-image-2.5-sunburst. OpenAI describes it as "built for premium visual workflows that benefit from tighter control across edits," naming production-ready campaign creative and polished product imagery. It takes longer to generate.
The operative phrase is across edits. Sunburst is not simply a higher quality setting. Its advantage shows up on the third and fourth revision of the same image, when everything you did not ask to change has to survive.
They cost the same
This surprises people, so it is worth stating plainly. Both models bill at identical rates:
| Token type | Rate per 1M tokens |
|---|---|
| Image output | $30.00 |
| Image input | $8.00 |
| Cached image input | $2.00 |
| Text input | $5.00 |
| Cached text input | $1.25 |
There is no price penalty for choosing Sunburst. What you pay for is time, not money.
What actually changes your bill
Not the model. The quality setting.
OpenAI bills by token, and a quality setting decides how many output tokens an image consumes. Both 2.5 models offer low, medium, high, xhigh, max and auto. The previous generation stopped at high.
Per image at 1024x1024, counting image output tokens only:
| Quality | GPT Image 2.5 | GPT Image 2 |
|---|---|---|
| low | $0.00588 | $0.00588 |
| medium | $0.01317 | $0.05268 |
| high | $0.05268 | $0.21072 |
| xhigh | $0.09366 | not available |
| max | $0.21072 | not available |
Read across the rows and the pattern is not "2.5 is cheaper". The labels moved. Medium on 2.5 costs a quarter of medium on GPT Image 2. High on 2.5 costs what medium used to cost. The new max costs what high used to cost. The old ceiling became the new middle, with two settings added above it.
Low to max on the same model is a thirty-six-fold difference in what one image costs to produce. That range is larger than any difference between Flare and Sunburst.
Larger canvases cost more. The same medium setting is $0.02676 at 2048x2048 and $0.02595 at 3840x2160.
How to choose
Pick Flare when you are generating at volume, when latency is visible to a user waiting on a screen, when you are prototyping and will throw most results away, or when you simply do not know yet. It is the default for a reason.
Pick Sunburst when the image is a deliverable rather than a draft, when a client or brand guide constrains what may change, or when you are on revision three of the same asset and drift has started to show.
Then pick the quality setting deliberately, because that is the decision that moves your bill. A medium Flare image and a max Sunburst image differ by roughly thirty-six times in cost. Choosing the model without choosing the setting is choosing nothing.
Questions worth asking a tool
If a product tells you it "upgraded to ChatGPT Images 2.5", it has told you very little. Three questions get you the rest:
- Which model, Flare or Sunburst? They behave differently across edits.
- Which quality setting? This is the whole cost story, and the honest answer is a word like
medium, not "high quality". - What happens on a failed generation? Technical failures should return your credits automatically.
We run a menu of named models rather than one hidden backend, and each entry shows its full name and its credit price per image. That is a direct response to the paragraph above.
Availability
ChatGPT Images 2.5 reached all ChatGPT tiers plus ChatGPT Work and Codex on desktop, mobile and web on release day. Both API models shipped the same day. OpenAI applies C2PA metadata and invisible watermarking to images made with its tools.
Sources: OpenAI, Introducing ChatGPT Images 2.5, and the OpenAI API pricing and image generation documentation, all read 9 September 2026. Per-image figures come from the calculator in OpenAI's own image generation guide.
ChatGPT, GPT Image and OpenAI are trademarks of OpenAI. Image 2.5 is our product name, not a model name. We are an independent workbench and not affiliated with OpenAI.