GPT Image 2.6
OpenAI's image model from 4 November 2026, called Images 2.6 in ChatGPT. Below: what changed, which variant to use, what it costs, and prompts you can copy, all illustrated with images we generated ourselves.
What's new in GPT Image 2.6
- Animated GIFs. A prompt or a photo becomes a looping GIF; speed and single frames can be edited afterwards. Full guide: AI GIF generator for ChatGPT.
- SVG output. One static vector file per request, suited to icons, logos and flat illustrations.
- Reference images that persist through a conversation, so a character or product stays the same across several images.
- Faster generation: OpenAI cites about 50% faster than the previous version.
Flare vs Sunburst
In the API GPT Image 2.6 comes in two variants. Flare is the faster model: use it for drafts, many variations and anything you will pick from. Sunburst is the full model: use it for the final image, dense text, and edits where faces, products or layouts must not drift. ChatGPT chooses for you.
GPT Image 2.6 API pricing
| Reported rate (both variants) | |
|---|---|
| Text input | $5 per 1M tokens |
| Image input | $8 per 1M tokens |
| Image output | $30 per 1M tokens |
| Batch | Not offered for 2.6 |
The rate card matches GPT Image 2.5, but low and medium quality images reportedly use about four times as many output tokens, so cheap drafts cost more than before while high quality costs about the same. These figures come from third-party analyses of OpenAI's calculator; confirm on OpenAI's pricing page.
GPT Image 2.6 vs GPT Image 2.5 vs Nano Banana
| GPT Image 2.6 | GPT Image 2.5 | Nano Banana (Gemini) | |
|---|---|---|---|
| Still images | Yes | Yes | Yes |
| Animated GIF | Native | No | No; assemble frames yourself |
| SVG file output | Yes, static | No | No |
| Batch API | No | Yes | — |
How to prompt GPT Image 2.6
- Format and shape first: "Generate an image, 3:4" or "Make a looping GIF, 1:1".
- Subject and composition: what is in the frame and where.
- Medium and palette: name a real medium (risograph, gouache, 16-bit pixel art) and three or four colours.
- Exact text in quotes, if any. Short titles render reliably.
- What must not change: camera, face, product. This matters most for GIFs and edits.
Generate an image, 3:4: a three-day Kyoto travel guide infographic in Chinese, illustrated map with numbered stops, small watercolour vignettes for Day 1/2/3, clean timeline, title '京都三日游攻略', cream paper.
Every example on mimus shows its full prompt; browse them all on GPT Image 2.6 prompts.
Questions
What is GPT Image 2.6?
GPT Image 2.6 is OpenAI's image model released on 4 November 2026, shown in ChatGPT as Images 2.6. It adds animated GIF output, SVG output, reusable reference images within a conversation and faster generation.
Is GPT Image 2.6 free in ChatGPT?
It is available in ChatGPT on the web, desktop and mobile; GIF output is rolling out gradually across plans including Free. Usage limits per plan are not published.
What is the difference between Flare and Sunburst?
They are the two API variants. Flare is the faster one for drafts and volume; Sunburst is the full model for detailed finals and edits that must stay consistent.
How much does the GPT Image 2.6 API cost?
Reported rates are the same for Flare and Sunburst: $5 per million text input tokens, $8 per million image input tokens and $30 per million image output tokens, with no Batch discount. Low-quality images reportedly use about four times the output tokens of GPT Image 2.5. Check OpenAI's pricing page before budgeting.
Can GPT Image 2.6 make GIFs and SVGs?
Yes. GIFs come from a prompt or a photo (the 14 we measured had 9 to 12 frames at 110 to 180 ms); SVG output is one static vector file per request, best for icons and simple illustrations.