GPT Image 2.5 — Developer API Guide

Release date: 2026-09-09 Applies to: Imaginer Developer API (all API plans)

Two new image generation models are now available on the Developer API: GPT Image 2.5 Sunburst and GPT Image 2.5 Flare. Both models support text-to-image and image-to-image (reference images), multiple aspect ratios, 21 style presets, and two quality tiers.

---

Model IDs

Model	model_id
GPT Image 2.5 Sunburst	gpt-image-2.5-sunburst
GPT Image 2.5 Flare	gpt-image-2.5-flare

Use these exact identifiers in the model_id field of POST /api/public/v1/generate.

---

Capabilities

Both models share the same capabilities:

Capability	Value
Quality tiers (quality)	low (default), More Better
Aspect ratios (ratio)	1:1, 2:3, 3:2, 16:9, 9:16
Resolution tier	Medium (e.g., 1:1 = 1024×1024, 2:3 = 848×1264, 3:2 = 1264×848, 16:9 = 1376×768, 9:16 = 768×1376)
Reference images (ref_image_ids)	Supported, up to 6 images
Styles (style)	21 presets (see list below)

Quality tiers

• low — base tier, fastest and cheapest. Used by default when quality is omitted.
• More Better — enhanced tier, higher fidelity at a higher price per image.

The quality value is case-insensitive and tolerant of separators — More Better, more better, MORE_BETTER, and more-better are all accepted. Any other value returns 400 with Invalid quality. Allowed: low, More Better.

Billing tier

On the Pricing page, the More Better tier is billed as the MEDIUM tier, and low as LOW.

Style presets

dynamic, cinematic, creative, fashion, portrait, portrait-cinematic, portrait-fashion, illustration, 3d-render, acrylic, game-concept, graphic-design-2d, graphic-design-3d, pro-b-w-photography, pro-color-photography, pro-film-photography, ray-traced, stock-photo, vibrant, watercolor, none

---

Example: Generate an Image

Endpoint: POST /api/public/v1/generate

POST /api/public/v1/generate
authorization: Bearer <API_KEY>
Content-Type: application/json

{
  "model_id": "gpt-image-2.5-sunburst",
  "prompt": "A serene Japanese garden at sunrise, soft morning mist",
  "ratio": "16:9",
  "quality": "More Better",
  "style": "cinematic"
}

Response (202 Accepted):

{
  "status": "processing",
  "generation_id": "uuid-string-here",
  "message": "Request accepted. Generation is processing in the background. Poll /generate/{id} for status."
}

Generation is asynchronous. Poll GET /api/public/v1/generate/{generation_id} until status is complete, then download the result via GET /api/public/v1/generate/{generation_id}/image/{index} (index starts at 0).

With reference images (image-to-image)

Upload each reference first via POST /api/public/v1/upload, then pass the returned IDs:

{
  "model_id": "gpt-image-2.5-flare",
  "prompt": "Place this product on a marble countertop, studio lighting",
  "ratio": "1:1",
  "quality": "low",
  "ref_image_ids": ["uuid-ref-1", "uuid-ref-2"]
}

---

Pricing & Rate Limits

• Each quality tier has its own price per image. The More Better tier costs more RPM budget, concurrency budget, and IDR credits than low. See the Pricing & Rate Limits page for the exact figures.
• The response headers X-RateLimit-Cost and X-Image-Price (credit plans) show the exact cost of each request.
• Current per-tier prices and limits are listed in your developer dashboard and may be adjusted over time.

---

Quick Reference

Minimal request — defaults to ratio 1:1, quality low:

{
  "model_id": "gpt-image-2.5-flare",
  "prompt": "A cat wearing sunglasses"
}

Field	Required	Values
model_id	Yes	gpt-image-2.5-sunburst, gpt-image-2.5-flare
prompt	Yes	Text, max 2000 characters
ratio	Optional	1:1 (default), 2:3, 3:2, 16:9, 9:16
quality	Optional	low (default), More Better
style	Optional	One of the 21 style presets
ref_image_ids	Optional	Up to 6 IDs from the upload endpoint