Forge is available: build a website with AI from your RodiumAi account.

Try Forge
OpenAI

GPT-Image-2.5 Sunburst

Healthyimage-genediting
Context
-
Max output
-
Speed
Balanced
Quality
Standard
Released
2026-09-08
Input: / Output:
ImageTextImage

About this model

Premium image generation and editing with precise control across iterative edits. GPT-Image-2.5 Sunburst is OpenAI's most capable model for image generation and editing, designed for premium workflows that need tighter control across detailed and iterative edits.

Pricing

RateRODIUSD (ref.)Unit
Image
Image out126.6~ 0.167per image (RODI/img, USD/img)

Pricing tiers

The rate depends on how you call the model. Each table below lists every tier, in RODI with the approximate USD value.

From 4.5 RODI (~ $0.005935) per image

Price per image by quality and size

Quality / size1024x1024
low4.5~ 0.005935
medium10.1~ 0.013225
high40~ 0.052775
auto4.5~ 0.005935
max159.7~ 0.210815

RODI prices include RodiumAi markup and upstream fees. USD figures are wholesale reference rates.

Providers & pricing

ProviderInputOutputCachedFree tierStatus
Runware

Runware

Default

All rates (1)
  • Image out126.6 RODI/img
----No data

Capabilities

Streaming
Tool calling
Vision
JSON mode
Reasoning

API usage

Generate images with client.images(): POST /v1/images/generations.

Examples

…

Example response

…

Reference image (edit or restyle)

…

Quality and several variants

…

RodiumAi SDK

images() maps to POST /v1/images/generations for this image model.

…

Use cases

  • Marketing visuals, social posts and ad variants from a short brief.
  • Product shots and mock-ups for a catalogue.
  • Illustrations for articles, slides and onboarding screens.
  • Concept art and moodboards before a design review.
  • Backgrounds, textures and placeholder images for apps.

Notes

  • GPT Image models are billed per token (text input, image input and image output), and the cost depends on quality and size. The pricing table above shows the rates.
  • quality accepts low, medium, high and auto. Use low for drafts and high for final assets.
  • Reference images for edits must be inline base64 (data URL or raw base64), not an https:// URL. Send one in image or up to 14 in images. With a mask it becomes an inpainting edit.
  • Images come back as b64_json.
  • Good at rendering readable text inside images (posters, labels, UI mock-ups).
View all