by OpenAI · Released Nov 4, 2026
Turn a prompt into a finished image in seconds — then point at whatever should change instead of describing it. 50% faster than 2.5, with GIF output, in-image markup editing and SVG export.
GPT Image 2.6 for generation and editing
Four things, per OpenAI's release notes: it generates 50% faster; it can turn a prompt or an image into an animated GIF (adjustable speed, editable frames); the API now supports image markup — you draw on the image to mark exactly what should change — plus SVG export; and it keeps characters, products and brand details consistent across edit rounds.
The 6-second claim has a condition, and most sites get it wrong: it applies to Flare at quality=low. Sunburst at medium or high takes longer — that is the whole point of having two models.
GPT Image 2.6 for generation and editing
Four things, per OpenAI's release notes: it generates 50% faster; it can turn a prompt or an image into an animated GIF (adjustable speed, editable frames); the API now supports image markup — you draw on the image to mark exactly what should change — plus SVG export; and it keeps characters, products and brand details consistent across edit rounds.
Describing "change the label on the box, keep the lighting" is where edit rounds usually break. With markup, you draw on the image and mark the region — the model works on that area and leaves the rest alone. The same capability is exposed in the API in 2.6, and finished work can go out as SVG for web, app and print.
2.6 animates from a prompt or an image. In practice the reliable pattern — and the one in OpenAI's own launch examples — is to ask for a sprite sheet: one grid of sequential frames, say 6 frames in 3 columns × 2 rows, read left to right then top to bottom, same character, same camera, identical panel size, and a final frame that flows back into the first. Then set the speed and edit individual frames. Craft rules that matter: fix the camera, make each frame a small increment, keep the last frame adjacent to the first, and ask for no captions, no borders, no speech bubbles.
The most expensive part of image work isn't the first generation — it's round four, when the face, the logo and the label have drifted. 2.6 is built for that: save a reference image once (a person, a pet, a character, a product), reuse it across new creations, and edit in passes instead of re-prompting from scratch.
Microsoft has a different model with a confusingly similar name, MAI-Image-2.6. This page is about OpenAI's GPT Image 2.6 — the model released on November 4, 2026, available in ChatGPT, Codex and the API. If you came looking for Microsoft's model, you're on the wrong page; if you want OpenAI's, you're in the right place.
How to get started
Our recipe: lock the composition at low complexity first, then raise quality for the final render. You spend the expensive tokens once, on the version that survives.
quality controls detail and fidelity.
Use low complexity for most editing tasks, and reach for medium or high only when a scene has many elements or constraints.
2.6 ships as two API models — gpt-image-2.6-flare and gpt-image-2.6-sunburst — and they are tuned differently.
OpenAI: as little as 6 seconds at low
GPT Image 2.6 and 2.5
Didn't change: Flare and Sunburst are not new in 2.6. OpenAI's price list carries gpt-image-2.5-flare and gpt-image-2.5-sunburst at the same per-token rates as the 2.6 pair — $8 per 1M image input tokens, $30 per 1M output tokens, $2 cached. Switching models doesn't change token price; your cost difference comes from how many output tokens a setting actually produces.
Start free with 20 credits — no card required. Free tier is personal use only, includes standard generation speed, keeps history for 7 days, and is limited to 1 image per batch job. Paid plans start at $19.99/month ($11.99 billed yearly, 500 credits), with Pro at $89.99 ($23.99 yearly, 1,600 credits, up to 4K) and Premium at $149.99 ($59.99 yearly, 4,000 credits). Commercial licensing comes with qualifying paid plans. See the full plan comparison on the pricing page.
FAQ
How much it costs, which model to pick, and what you can do with it.
Yes to start: 20 free credits on sign-up, no card required. The free tier is for personal use, at standard generation speed, with 7-day history and 1 image per batch job.
Start with 20 credits — generate from a description or edit an existing photo, and keep refining until it's ready.
