Skip to main content
POST
Since September 10, 2026, gpt-image-2 and gpt-image-2-hd have been renamed to gpt-image-2.5 and gpt-image-2.5-hd. The old model IDs are no longer accepted. Update the model in your requests; these are not compatibility aliases.
The base prices for these per-call models are USD 0.0441/call and USD 0.005479/call, respectively; group multipliers may affect the final charge. This rename does not change prices. Read image results when data is returned. An image.generation.task response means an asynchronous task was submitted, not that an image is ready. The response shape depends on the route; do not treat task acceptance as the final result. Flare and Sunburst are separate token-billed models. They are not part of this per-call contract and are not yet available.

Reference image limits

The reference-image field is images (an array of image URLs or base64 strings); image, image_urls and input_reference are accepted aliases and are normalized to images.
  • gpt-image-2.5: up to 16 images per request; no minimum (omit images for text-to-image); the vendor publishes no accepted-format list or per-file size limit for reference images.
  • gpt-image-2.5-hd: no model spec exists yet, so its reference-image limits are unverified and no image count, format or per-file size limit is stated here.
  • gpt-image-2.5-flare: discounted tier image model route; reference image specifications follow baseline.
  • gpt-image-2.5-sunburst: economy tier image model route; reference image specifications follow baseline.
The 50 MB figure in the vendor’s image guide is scoped to the image being edited together with its mask, not to a single reference image; no per-file ceiling is stated here.

Authorizations

Authorization
string
header
required

Bearer authentication header of the form Bearer <token>, where <token> is your auth token.

Body

application/json
model
enum<string>
default:gpt-image-2.5
required
Available options:
gpt-image-2.5,
gpt-image-2.5-hd,
gpt-image-2.5-flare,
gpt-image-2.5-sunburst
prompt
string
required
n
integer

Image count, subject to the selected model and route limits.

size
string

Accepted sizes depend on the selected model and route; not every size is guaranteed.

quality
string

Quality options depend on the selected model and route.

images
string[]

Reference images: image URL or base64. image / image_urls / input_reference are accepted aliases and are normalized to images.

Response

Image results or an accepted asynchronous task.

The response can contain image results or an asynchronous task; handle the returned shape.

created
integer
data
object[]
id
string

Asynchronous task ID, when returned.

object
string
model
string
status
string