GPT Image 2.6 API Guide: Models, Pricing and Parameters
Updated November 5, 2026 · 6 min read
GPT Image 2.6 is available in the OpenAI API as two models, GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst. This guide collects what is known at launch: endpoints, output formats, the new parameters and pricing. Details may change, so confirm against OpenAI's official documentation before you ship.
Models
| Model | Tuned for | Notes |
|---|---|---|
| GPT-Image-2.6 Flare | Latency | As fast as ~6 s per image at low quality |
| GPT-Image-2.6 Sunburst | Quality and consistency | Best at medium/high quality; reported ID gpt-image-2.6-sunburst |
Endpoints
POST /v1/images/generations— create an image from a promptPOST /v1/images/edits— edit with a mask or with markup
Inputs are text and images; outputs are images. The model does not accept video and is not a general chat model.
Output Formats
PNG, JPEG and WebP as before, plus two new formats: GIF for short animated loops and SVG for vectors. Early reports describe GIFs from 2 to 64 frames, subject to size and pixel limits, and GIF is not accepted as an input format. Reports differ on whether GIF and SVG are available from the edit endpoint, so test this with your own key.
The Complexity Parameter
complexity (low, medium, high) controls how much planning the model does before drawing — not the visual quality. OpenAI recommends starting with low complexity for fast edits and moving to medium or high for crowded scenes with many constraints. Combine it with quality, which controls detail and fidelity.
Markup-Guided Editing
To edit with markup, send the source image plus a transparent PNG overlay with the same dimensions where you have drawn the changes. Markup cannot be combined with a mask in the same request.
Pricing
Reported OpenAI list rates are the same for both models (per 1 million tokens):
| Token type | Price |
|---|---|
| Text input | $5.00 (cached $1.25) |
| Image input | $8.00 (cached $2.00) |
| Image output | $30.00 |
In practice that works out to roughly $0.005 for a 1K image at low quality, about $0.05 at high quality and about $0.10 for a 2K high-quality image. GIFs use more output tokens because they contain several frames.
A Practical Workflow
- Draft with Flare at low quality and low complexity.
- Lock the composition, then regenerate with Sunburst at medium or high quality.
- Use markup edits for local fixes instead of regenerating the whole image.
Don't want to manage keys and token bills? The GPT Image 2.6 generator on this site wraps both models in simple monthly credits — see pricing.