GPT Image 2.6

GPT Image 2.6 is OpenAI’s series released November 4: the same two tiers as 2.5, roughly 50% faster to generate. gpt-image-2.6-flare — speed and interactive iteration; gpt-image-2.6-sunburst — dense compositions, text-heavy layouts and edits that stay consistent from one iteration to the next. Both work with POST /v1/images/generations and POST /v1/images/edits.

The 2.6 series highlights in Mixen: animated GIF (512×512, 9 frames at 200 ms, a ready loop for stickers and previews) and vector SVG (infinite scaling, size chosen by the model). Both formats are generation-only (no references) and arrive as files in their native form.

In the web cabinet and the bot there is no format switch — just say it in the prompt: “make a GIF of a bouncing ball” or “logo in vector” — the format picks itself. Explicit control (output_format) remains in the public API.

Parameters

Field	Description
model	openai/gpt-image-2.6-flare or openai/gpt-image-2.6-sunburst (short gpt-image-2.6-flare / gpt-image-2.6-sunburst)
prompt	What to draw, required; up to 32,000 characters
n	How many images, 1 to 4; each is billed separately
aspect_ratio	Ratios from 1:3 to 3:1, including 2:1, 1:2, 3:1, 1:3, 21:9; defaults to 1:1
resolution	1k, 2k or 4k; defaults to 1k
size	Explicit WxH (both edges multiples of 16, long side up to 3840)
background	transparent | opaque | auto — the series’ background knob
output_format	png | jpeg | webp | gif | svg; defaults to png. GIF — 512×512, 9 frames × 200 ms (fixed); SVG — auto size; both generation-only
complexity	low | medium | high — planning depth (default high); low for targeted tasks. Only with output_format: gif/svg
output_compression	0–100 — JPEG/WebP compression (with png returns 400)
quality	The series has no such knob: passing it returns 400

Transparent background

background: "transparent" cuts the object out with no backdrop — the output format must then be png or webp, otherwise 400. auto lets the model decide, opaque forces a filled background.

curl https://api.mixen.ai/v1/images/generations \
  -H "Authorization: Bearer $MIXEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2.6-flare",
    "prompt": "Sticker: capybara astronaut, outline style",
    "aspect_ratio": "1:1",
    "background": "transparent",
    "output_format": "png"
  }'

Editing from references — up to 16 images

/v1/images/edits takes up to 16 references per request:

curl https://api.mixen.ai/v1/images/edits \
  -H "Authorization: Bearer $MIXEN_API_KEY" \
  -F model=openai/gpt-image-2.6-sunburst \
  -F image=@model.png \
  -F image=@jacket.png \
  -F image=@studio-light.png \
  -F prompt="Dress the model in this jacket, light as in the third reference"

Text-to-image

curl https://api.mixen.ai/v1/images/generations \
  -H "Authorization: Bearer $MIXEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2.6-flare",
    "prompt": "Final advertising shot: perfume on a marble counter, water drops, side light, photorealism",
    "aspect_ratio": "3:1",
    "resolution": "4k"
  }'

Price

Raster formats — 3.90 ₽ per image for both tiers. GIF and SVG are priced separately (the official route bills per token: frames and detail); the charge follows the actual generation, with the pre-check shown before the run. Current values — catalog.

Animated GIF and SVG

curl https://api.mixen.ai/v1/images/generations \
  -H "Authorization: Bearer $MIXEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2.6-flare",
    "prompt": "A ball bouncing across a table, seamless loop",
    "output_format": "gif"
  }'

• GIF: 512×512, 9 frames × 200 ms (~1.8 s loop) — parameters are fixed by the service
• SVG: the model picks the size (you can omit size); ideal for icons and logos; background: "transparent" is supported — alpha is native to SVG
• both formats: POST /v1/images/generations only — they do not combine with references (/v1/images/edits), use png/jpeg/webp there

Tip for a seamless loop (OpenAI guide): give the subject a pose it returns to (“cat crouches — leaps — crouches again”) and pin the camera (“camera still, background unchanged”) — the first and last frames then match.

Strengths and limitations

• Up to 50% faster than GPT Image 2.5; two tiers per task: Flare — speed and iteration, Sunburst — detail and edit consistency.
• Native transparent backgrounds (background + png/webp).
• Up to 16 references in /v1/images/edits; prompts up to 20,000 characters.
• Frame ratios from 1:3 to 3:1, sizes up to 4K.
• GIF/SVG are generation-only; with references or a transparent background the service returns png instead.
• No quality knob; background/output_format sent to other models return 400.
• Balance-only — not part of the free quota.

All models — in the catalog. The general image workflow — images guide.