GPT-Image-2.5
OpenAI’s image model, shipped as two separate API models: gpt-image-2.5-flare and gpt-image-2.5-sunburst.
- Two GA models
- Transparent backgrounds
- Mask-based inpainting
- Up to 16 reference images
- Streaming previews
OpenAI released GPT-Image-2.5 on 8 September 2026. The name covers two models rather than one: the API exposes gpt-image-2.5-flare and gpt-image-2.5-sunburst, and choosing between them is the first decision anyone integrating it has to make.
Both are served from the same two image endpoints as GPT-Image-2 and take the same parameters, with two new quality settings on top. Both are also in the Labnana generator, where they cost the same per generation as GPT-Image-2 does on the discount channel.
What changed since GPT-Image-2
2.5 is served from the same two image endpoints as GPT-Image-2, plus the Responses image tool. The additions below are what a working integration has to account for before it switches the model string.
Two quality settings above the old ceiling
Earlier GPT Image models stop at high. Both 2.5 models add xhigh and max on top, and both still default to auto. OpenAI states that earlier models do not support these two values, so shared code has to pick the setting per model rather than pass one through.
Transparent backgrounds are documented per model
The background parameter takes opaque or transparent on both 2.5 models. Transparency needs a PNG or WebP output format, because a JPEG has no alpha channel for the setting to write into.
Editing: masks, references and chained turns
An edit request takes up to 16 input images and an optional mask, so a change can be scoped to part of the picture. Through the Responses image tool an edit can also chain off the previous response instead of restating the whole scene.
Same rates, unknown token counts
Per-token pricing is identical to GPT-Image-2. How many tokens an image consumes is a separate question: OpenAI states that the GPT-Image-2 calculator does not estimate GPT-Image-2.5 consumption, so the new quality tiers have no published cost per image.
Flare or Sunburst
Both models bill at the same rates, so the choice is about output rather than cost.
gpt-image-2.5-flare
OpenAI’s models index describes it as “Fast, high-quality everyday image generation”, and the image generation guide says to choose Flare for exactly that. It is the default of the two.
gpt-image-2.5-sunburst
The models index calls it “Our most capable model for image generation and editing”, and the guide says to choose Sunburst “for workflows where editing precision matters most”.
Dated snapshots
Both models publish a 2026-09-08 snapshot alongside the rolling alias. Pin the snapshot when output has to stay reproducible, and take the alias when you would rather inherit improvements.
Specifications
Read off OpenAI’s model pages, API reference and pricing page. Rates are per million tokens; there is no per-image price published for either model.
| Provider | OpenAI |
|---|---|
| Model IDs | gpt-image-2.5-flare · gpt-image-2.5-sunburst |
| Dated snapshots | Both models publish a 2026-09-08 snapshot |
| Endpoints | Image generation and image edit; also the Responses image tool |
| Quality | auto · low · medium · high · xhigh · max |
| Background | opaque or transparent (PNG or WebP output) |
| Reference images | Up to 16 per edit request |
| Output size | Up to 3840px on the long edge |
| Aspect ratio | From 1:3 to 3:1 |
| Editing | Mask-based inpainting; edits can chain turns |
| Streaming | Up to 3 partial images before the final one |
| Text input | $5.00 per 1M tokens |
| Image input | $8.00 per 1M tokens |
| Image output | $30.00 per 1M tokens |
| Cost per image | Not published — the GPT-Image-2 calculator does not apply |
GPT-Image-2.5 on Labnana
Both models are in the generator. Pick GPT-Image-2.5 Flare or GPT-Image-2.5 Sunburst from the model menu and generate at 1K, 2K or 4K with any of the supported aspect ratios. They run on the same discount channel as GPT-Image-2 and cost the same per generation, which the generate button shows live before you commit.
Which of the two to pick
They bill identically, so it is not a price decision. Flare is OpenAI’s everyday choice and the one to start from. Sunburst is the one they position above it for generation and editing. Run the same prompt on both and keep the one you prefer.
What it costs
The same per-generation price as GPT-Image-2 on the discount channel, by resolution. The generate button shows the exact figure for your current settings, so there is no separate price list to keep in sync.
Where it sits next to the rest
Nano Banana Pro is still the one to reach for when the image must contain readable text or look photographic, and it carries a standing free allowance. Seedream 5.0 Pro is the editing model: pin a spot on an image you already made and write what changes there.
GPT-Image-2.5 FAQ
Try GPT-Image-2.5 on Labnana
Flare and Sunburst are both in the model menu, alongside GPT-Image-2, Nano Banana Pro and Seedream 5.0 Pro. The same prompt runs on any of them.