Changelog

GPT Image 2.6 Flare and Sunburst are here.

A pale man with a blonde pageboy haircut and waxed handlebar mustache, in a green tweed three-piece suit, sits in a leather armchair in a wood-panelled library.
GPT Image 2.6 Sunburst · 1024 × 1024 · $0.0269A deadpan, close-up portrait of a blonde very pale man looking directly into the camera with an intense, unblinking expression. He has a prominent monobrow, a highly styled, sharply waxed upturned handlebar mustache, and a rigid, jet-blonde French pageboy bob haircut shaped perfectly like a solid mushroom helmet. He is wearing a crisp, high-collared tweed three piece suit…Made in Image Studio. Prompt shortened.

OpenAI’s GPT Image 2.6 is on MiniRouter in two versions, Flare and Sunburst. Write a prompt in Image Studio or call POST /v1/images/generations, and get a 1024 × 1024 image for about $0.0281 each.

Flare or Sunburst?

OpenAI describes Flare as made for interactive workflows and Sunburst as its general image model. Both take the same request and cost the same, so run your prompt on each and keep the one you prefer.

Model

OpenAI’s general GPT Image 2.6 model for creating images from a prompt.

$0.0281estimated per image

Output tokens per 1024 × 1024 image
2.6 Sunburst824
2.6 Flare824
GPT Image 2196
2.6 Sunburst · cURL
curl https://api.minirouter.sh/v1/images/generations \
  -H "Authorization: Bearer $MINIROUTER_KEY" \
  -H 'Content-Type: application/json' \
  --data '{
  "model": "openai/gpt-image-2.6-sunburst",
  "prompt": "A deadpan portrait of a pale man with a blonde pageboy haircut and a waxed handlebar mustache, green tweed three-piece suit, leather armchair, warm library light",
  "n": 1,
  "output_format": "png",
  "quality": "low",
  "size": "1024x1024"
}'

In our test requests, each image used 824 output tokens on either version, against 196 on GPT Image 2. The token rates are the same, $5 in and $30 out per million, so GPT Image 2.6 costs more per image. Use GPT Image 2 for cheap drafts.

Model pages: Sunburst and Flare.

What you pay.

Image Studio shows the estimate before you generate. Each image settles at OpenAI’s cost plus MiniRouter’s 5% fee, never more than the credit held while it runs. A request refused before it reaches OpenAI costs nothing.

Know the limits.

  • The example uses 1024 × 1024, low quality, PNG. Open the model in Image Studio to see its available sizes, quality settings and current estimate.
  • One image per request. Image Studio makes up to 4 by sending one request per image.
  • Text prompts only, so no reference images or edits yet. To edit a photo, use GPT Image 2.5 Flare or Sunburst.
  • The acceptable use policy prohibits deepfakes and intimate images of real people.

See what else shipped.