Skip to content
A woman under a clear umbrella on a rainy night street, made from one photo with the Rainy Night Paparazzi template on GPT Image 2.6 Flare
Made in Phosphene with GPT Image 2.6 Flare

What GPT Image 2.6 is good at

GPT Image 2.6 is OpenAI’s newest image model and the default model in Phosphene since 5 November 2026. It keeps the strengths of GPT Image 2.5, prompt adherence and exact text, and adds animated GIF output. Flare is the quick version; Sunburst is for the final asset.

Strong at

  • Following long, specific prompts closely
  • Exact text on packaging, posters and UI mockups
  • Editing a reference photo while keeping the face, product and layout
  • Short animated GIFs from a text prompt, 6 to 16 frames

Keep in mind

  • Flare takes about 12 seconds per image and Sunburst about 32; draft on Flare
  • Uses about three times the output tokens of GPT Image 2.5 at Medium quality, so it is priced above 2.5 once the Flare launch price ends on 19 November 2026
  • GIFs start from text alone; reference photos work for still images
  • Five aspect ratios only (square, 3:4, 4:3, 9:16, 16:9)

Tips for GPT Image 2.6

  1. Put the exact words you want in quotes, like "NORTH ROAST", so they are spelled as written.
  2. Draft on Flare at Medium quality, then switch to Sunburst or High quality for the final image.
  3. For a GIF, describe one simple motion that loops, like steam rising from a cup or a sign flickering on.

Versions in Phosphene

  • GPT Image 2.6 Flare

    OpenAI’s newest fast image model: generation and reference-photo editing for typography, product mockups, and complex prompts.

    From 2 credits
  • GPT Image 2.6 Sunburst

    OpenAI’s newest precise image model. Slower than Flare, built for finished assets and edits that must keep faces, products and layout intact.

    From 2 credits

GPT Image 2.6 specs

FlareSunburst
Price per imageFrom 2 creditsFrom 2 credits
Resolution1K, 1.2K, 1.5K, 4K1K, 1.2K, 1.5K, 4K
Aspect ratios1:1, 3:4, 4:3, 9:16, 16:91:1, 3:4, 4:3, 9:16, 16:9
Quality settingLow, Medium, HighLow, Medium, High
Reference photosYesYes
Typical timeAbout 12 secondsAbout 32 seconds

GPT Image 2.6 price per image

Credits for one image from a prompt, at each setting the composer offers.

Flare and Sunburst

ResolutionLowMediumHigh
1K2–33–54–6
1.2K2–43–64–7
1.5K2–43–74–8
4K6–99–1611–19

One reference photo adds up to 2 credits; more photos add up to 15 in total.

A range means the price depends on the aspect ratio. The composer shows the exact price before you generate.

Questions about GPT Image 2.6

How much does GPT Image 2.6 cost in Phosphene?
GPT Image 2.6 Flare and Sunburst: from 2 credits at 1K Low quality to 19 at 4K High. One reference photo adds up to 2 credits; more photos add up to 15 in total. The composer shows the exact price before you generate.
What is new in GPT Image 2.6 compared with GPT Image 2.5?
Animated GIF output. At Medium quality 2.6 also spends about 3.3 times the output tokens of 2.5 per image, which is why it is priced higher. Until 19 November 2026, Flare costs the same as GPT Image 2.5 Flare. The controls are the same, so prompts written for 2.5 work as they are.
What is the difference between Flare and Sunburst?
Flare is the fast version, about 12 seconds per image, and the place to start. Sunburst takes about 32 seconds; use it for final images and edits that must keep a face, product or layout intact.
How much does GPT Image 2.6 cost?
The price in credits is shown on this page and in the composer before you generate. Until 19 November 2026, Flare costs the same as GPT Image 2.5 Flare; after that it is priced on its own measured cost.
Can GPT Image 2.6 make animated GIFs?
Yes. Switch the composer to GIF, pick 6, 9 or 16 frames and describe the motion. GIFs are made from a text prompt and come out as small looping clips.
Do I need an OpenAI account or API key?
Phosphene runs the model for you and charges credits from your Phosphene balance, so a Phosphene account is all you need.

Other image models

See GPT Image 2.6 next to every model on the same prompts Compare all models