Lo esencial (BLUF): GPT Image 2.6 is the newest model of the family for creating and editing images: it draws a frame from a description and carefully edits your finished pictures. In NeuralSpace the model lives in the Images section, the public API and MCP. Three variants for different jobs: fast Flare, detailed Sunburst and budget Economy for bulk generation.
Flare is the fast variant for iterating on ideas: the image arrives sooner and the frame layout comes out tidier — mock-ups, covers, quick sketches. Sunburst is the quality variant: finer detail and steadier edit series when every element in the frame matters. Economy is the cheapest generation option: for when you need many images and quality is not critical.
Flare and Sunburst work both ways — as a text-to-image generator and as an editor of your own images. Economy handles text-to-image only and always renders at 1K.
The model outputs 1K, 2K and 4K images and understands 15 aspect ratios — from the 1:1 square to widescreen 21:9 and vertical 9:21. Output formats: PNG, JPEG, WEBP and GIF.
The background can be transparent: transparent, opaque and auto modes are available at 1K — handy for logos, stickers and collages that sit on any backdrop.
Upload up to 16 reference files (PNG, JPEG, WEBP, up to 30 MB each) and describe the changes in the prompt field — the model applies the edits while keeping the style of the source. It works for small touch-ups and for moving objects between frames alike.
Generation is billed in tokens: the exact price is shown before the run and stays frozen — it does not change after you press the button. Tokens are charged only for successful generations.
For your own apps the model is available in the public API as gpt-image-26: the request accepts variant, resolution, aspect_ratio and background. Through MCP the same model is available to AI agents. Create a key on the API page.
It is the next generation of the same family: 2.6 has three variants instead of two — the budget Economy joined Flare and Sunburst, and Flare composes the frame more neatly. Reference edits are steadier.
Flare when speed and a tidy layout matter, Sunburst for maximum detail and predictable edits, Economy for bulk generation at the lowest price (1K only).
Yes: at 1K resolution the background modes transparent, opaque and auto are available — the image is saved with a transparent backdrop in PNG or WEBP.
Up to 16 reference files, each up to 30 MB, in PNG, JPEG or WEBP. The model applies the changes and keeps the style of the sources.
Yes, in the public API as gpt-image-26 with the parameters variant, resolution, aspect_ratio and background; the model list and prices are on the API page.