Skip to main content
GPT Image 2.5 Flare and Sunburst launched on September 9, 2026 through official API forwarding, at the same token rates as GPT Image 2. Two per-call VIP models in the default group cost $0.03 per call.
Correction (September 16, 2026): web version 2.5 uses the model ID gpt-image-2.5-web. gpt-image-2-web is a legacy web ID used by existing integrations. The two IDs are separate routes, so do not mix them or reuse parameters between them.
  • Published: September 9, 2026
  • Last verified: September 24, 2026
  • Status: Active. Both -vip models were paused on September 16 and restored on September 22; see the recovery notice.
September 10 update: gpt-image-2.5-flare-vip and gpt-image-2.5-sunburst-vip support the xhigh and max quality tiers. Model IDs, groups, and per-call billing are unchanged.

​
Choose a model and route

Both Flare and Sunburst generate and edit images. Choose Flare for speed, Sunburst for edit precision, or offer both in your app.
  • Official forwarding: uses the existing GPT Image 2 official-forward integration and the Sora2Official or GPTImage2 Sora2 Enterprise group, with token billing at the same unit rates as official-forward gpt-image-2.
  • Per-call: uses the default group and model IDs ending in -vip.
  • Prices: per-call prices and group multipliers are listed in the model and pricing catalog and the console.

​
What users should do

  • Existing GPT Image 2 official-forward integrations: confirm that your token permits the target model, keep the Images API endpoint and authentication, and change model. Use POST /v1/images/generations for generation and POST /v1/images/edits for editing.
  • Existing per-call integrations: keep the per-call token, endpoint, and request structure used for gpt-image-2-vip, then change model to gpt-image-2.5-flare-vip or gpt-image-2.5-sunburst-vip.
  • Web-route integrations: for web version 2.5, set model to gpt-image-2.5-web. If your integration uses the legacy gpt-image-2-web, test gpt-image-2.5-web with your own parameters before switching, because the two routes do not share parameters.
  • Other existing routes: this notice does not require a universal migration. To adopt 2.5 official forwarding, verify the model, group, and usage-based billing settings.
Find generation, editing, image-saving examples, and size options in the GPT Image 2.5 integration guide. After a request, view the model, usage, and charge in call logs.

​
Frequently asked questions

​
Do matching GPT Image 2 rates mean the same cost per image?

No. The official-forward token rates match, but the model, quality, size, and reference images change token consumption and the total price. See the billing guidance and current console pricing; contact support for enterprise quotes.

​
Can the default group call Flare or Sunburst directly?

No. Unsuffixed Flare and Sunburst require a usage-based token in an official-forward group. The default group uses the per-call models gpt-image-2.5-flare-vip and gpt-image-2.5-sunburst-vip, and its web version 2.5 route is gpt-image-2.5-web.

​
Do the per-call models support only six sizes?

No. Both per-call models support standard 1K, 2K, and 4K sizes in common square, landscape, and portrait formats, plus custom dimensions within the valid size range; the guide lists six common examples.
  • size: set it as WIDTHxHEIGHT.
  • quality: choose one of these values:
    • low
    • medium
    • high
    • xhigh
    • max
  • To start, use size="2048x2048" and quality="medium", and omit any field you do not need.
See size and quality parameters for bounds and examples.

​
Can the per-call models make transparent images or remove backgrounds?

Yes. Both gpt-image-2.5-flare-vip and gpt-image-2.5-sunburst-vip support transparent PNG generation and background removal:
  • Set background="transparent" and output_format="png".
  • Use /v1/images/generations to generate a transparent image.
  • Use /v1/images/edits to upload an image and remove its background.
See the transparent PNG integration guide for complete examples.

​
Can I pin a model version?

For official forwarding, set model to a dated snapshot ID from the table. Per-call models use IDs ending in -vip, and web version 2.5 uses gpt-image-2.5-web; neither uses dated snapshot IDs.