ChatGPT Images 2.6: GIFs, Reference Images, and New API Pricing

November 6, 2026 · AI Radar

On November 4, OpenAI upgraded ChatGPT's image generation to Images 2.6. The headline making the rounds is "it can make GIFs," but the part that's more interesting for a business isn't in the headline: you can use a reference image you've saved without uploading it again every time.

Think of a product photo, a mascot or your own portrait. Dragging the same image into the chat for every new post is friction. That friction goes away. The GIF part is more for fun: a short greeting, an announcement card with a single animation. It can be useful; it won't change your life.

Short answer: Images 2.6 is now available on all plans in ChatGPT and Codex; it generates animated GIFs from text or a photo and remembers the reference images you save. On the API side, there are two new models (Flare and Sunburst), and pricing is the same as 2.5. For a small business (SMB) doing visual work, the most concrete gain is the reference image feature; trying GIFs costs nothing, and there's no need to rush to the API.

What changed?

According to OpenAI's announcement, four things stand out.

GIFs. It generates looping animated GIFs from a text prompt or an image you upload. You can adjust the speed and edit individual frames. The limits published on the developer side (2 to 64 frames, 20 to 2,000 milliseconds per frame) come from a third-party news site [C]; OpenAI's announcement doesn't officially state a GIF length or frame limit.

Reference images. You can save and label images of yourself, your pet or a character, then call them up in later generations without uploading them again. That makes it easier to keep the same character consistent across different scenes. How consistent it actually stays hasn't been independently measured, and OpenAI doesn't give a figure either.

Speed. The company talks about roughly 50% faster generation. For the fast model, Flare, it gives a figure along the lines of "six seconds at low quality." That's a vendor claim [A], with no independent measurement.

API. Two new models: GPT-Image-2.6 Flare (focused on speed) and GPT-Image-2.6 Sunburst (focused on quality and consistency across edits). Also new: editing by drawing marks on the image (markup), SVG export, and a "complexity" setting that determines how much the model plans before generating.

OpenAI says more than 4 billion images are generated each week; eight weeks ago, when 2.5 launched, that figure was 3 billion. The growth looks real, but this is also the company's own number.

Who is it for?

• Local shops and boutique brands that run their own social media: The reference image feature helps you most. You can save your product or your brand's mascot once and reuse it in your weekly posts.
• Agencies and freelance designers: Sunburst's promise of "consistency across edits" could help with client revisions. There's no evidence yet of it holding up in agencies' own work; don't make promises to a client before you've tried it.
• E-commerce sellers: SVG output is handy for simple icons and illustrations, but OpenAI states in its documentation that only one image can be requested in SVG and that editing isn't supported [C]. It's no substitute for catalog photos.
• If you don't generate images: You can skip it. This isn't text or automation news.

For images with Turkish text in them (posters, campaign cards), how cleanly 2.6 renders the letters is not clear in the official source. The best way to find out is a quick test with your own slogan.

How much does it cost?

Prices are from OpenAI's developer pricing page, as of November 6, 2026. The pricing page doesn't give a fixed amount per image; you work out the cost from the token usage in the response. On the forum, some developers write that at low and medium quality, 2.6 uses more output tokens for the same request [C]. OpenAI hasn't confirmed this. So even if the unit price is the same, your bill may go up; check the cost of a few images before you start even a small batch job.

There's no separate price for using it in ChatGPT. The feature is available on all plans, within your existing plan limits.

The one thing you can try today: In ChatGPT, save a product photo or your mascot as a reference image, then ask for two different scenes ("on a rainy street," "in a New Year's shop window"). If the same object is recognizable in both scenes, the feature works for you. It takes five minutes, and the free plan is enough.

If you want to speed up your own social media content workflow with AI, take a look at our AI automation page. In a free digital analysis, we can work out together which tasks are genuinely a good fit for tools like these.

AI Radar

I'm not including the company's own benchmark score for Inkling 2 Small here ([A], no independent verification); it's worth a look once the weights are open.

FAQ

Is ChatGPT Images 2.6 free?

According to OpenAI's announcement, it's available on all subscription tiers in ChatGPT and Codex. Daily usage limits per plan apply; the official announcement doesn't make the numbers for those limits clear.

Can I use the GIFs on social media?

Technically, yes: you can download them as GIF files and share them. The announcement doesn't separately address commercial use terms; check OpenAI's terms of use before using them in a brand campaign.

Do I need to switch to the API?

Not if you don't generate images with code. The ChatGPT interface has all of the new features. The API is only for people who want to add image generation to their own app or automation.

Sources

• OpenAI — Introducing Images 2.6 [A]
• OpenAI developer pricing page [A]
• OpenAI Developer Community — announcement thread [C]
• Storyboard18 — Images 2.6 coverage [C]
• ETV Bharat — Images 2.6 [C]
• AdGraam — Images 2.6 pricing and access [C]
• Thinking Machines — Inkling 2 Small [A]
• OpenRouter — tencent/hy4 [C]