ChatGPT Image 2.6: What's New in OpenAI's Images 2.6 Update

“ChatGPT Image 2.6” refers to ChatGPT Images 2.6, the image-generation update OpenAI released on November 4, 2026. It adds animated GIF creation, saved reference images, markup-guided edits and SVG export, and it is available on every ChatGPT plan, in Codex, and through two new API models called GPT-Image-2.6 Flare and Sunburst.

Below is what actually changed, what is still unclear, and what developers should know about cost. Everything here reflects what OpenAI and reputable outlets reported as of November 6, 2026; details may change as the rollout continues.

What’s new in ChatGPT Images 2.6

OpenAI’s launch post summed it up as “Now with GIFs and reference images,” plus two API models with “50% faster image generation.” Here are the headline features.

Animated GIFs

You can now turn a prompt or an image into an animated GIF, adjust its playback speed, edit individual frames, and share it as a looping animation. Coverage from heise also notes you can refine a finished GIF with follow-up text prompts.

One important caveat: the GIF feature is rolling out gradually. Images 2.6 itself is live, but not every account sees GIF creation yet.

Reference images

You can save images of yourself, your pets or recurring characters and reuse them in new creations without uploading them every time. This matters most for consistency, for example keeping the same character across a series of images or across the frames of a GIF.

A practical privacy note raised in some coverage: if you use a personal Free, Go, Plus or Pro account, think twice before saving photos of colleagues or customers, since conversations on those plans can be used for model training unless you turn that setting off.

Markup and SVG export

Images 2.6 also brings image markup, letting you highlight the exact area of an image you want changed rather than describing it in words, and SVG export, which gives you vector output that scales cleanly for logos, icons and simple graphics. Both are reported by heise and other outlets. We did not find an official step-by-step walkthrough for markup in the ChatGPT app at the time of writing.

Speed

OpenAI claims 50% faster image generation for the new API models. Be aware that the comparison baseline is not clearly stated in the announcement, and some coverage has mixed up this claim with similar wording from the earlier Images 2.5 launch, so treat “50%” as OpenAI’s headline figure rather than a guaranteed number for your own prompts.

Who can use it and where

According to OpenAI, Images 2.6 is available “in ChatGPT, and Codex across all tiers on web, desktop, and mobile,” and the Flare and Sunburst models are available in the API.

What OpenAI has not published yet:

• Maximum GIF length or frame count
• How many images or GIFs each plan allows
• Whether a GIF counts as one image or several toward your usage limit
• Whether the Free plan has different limits for GIFs and reference images

If you hit a limit, ChatGPT’s usage messages and OpenAI’s help center are the authoritative place to check.

How to try Images 2.6 in ChatGPT

OpenAI hasn’t published a detailed in-app guide, but the update works inside the normal ChatGPT image flow:

1. Update the ChatGPT app (or refresh the web version) so you are on the latest release.
2. Start a chat and describe the image you want, or upload an image you want to edit.
3. To animate it, ask ChatGPT to turn the prompt or image into a GIF. If nothing happens, the GIF feature may not have reached your account yet.
4. Use follow-up prompts to change the speed or adjust specific frames.
5. For consistent subjects, save a reference image of the person, pet or character and refer back to it in later chats.
6. Download the result in the format you need, including SVG where the option is offered.

Only use the official ChatGPT apps and website. Third-party sites advertising “Images 2.6” access are not OpenAI products, even if they use the API under the hood.

Flare vs. Sunburst: the API models

Images 2.6 ships to developers as two models. Each has its own page in OpenAI’s docs: GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst.

The Flare/Sunburst naming carried over from Images 2.5, and coverage reports the token rates are unchanged from that generation. Because pricing is per token rather than per image, your real cost depends on resolution, quality setting and how many edits you run. Some independent analyses also say SVG output is billed partly at the image model’s rate and partly at the rate of the model that writes the SVG; that detail comes from third-party blogs, so confirm it on OpenAI’s pricing page before budgeting.

Which endpoints each model supports has been reported inconsistently, so check the model pages directly before migrating existing code.

Safety and provenance

OpenAI published a system card for Images 2.6. Key points:

• The safety stack is built on the same foundations as Images 2.5, with added safeguards specific to GIFs.
• OpenAI added synchronous CSAM classification alongside its existing image moderation.
• Generated images carry C2PA provenance metadata plus SynthID, an invisible watermark that works alongside the metadata. In practice, that means images can be identified as AI-generated even if someone strips the file’s metadata, though no watermarking method is foolproof.

The system card also acknowledges that moderating motion in GIFs is harder than moderating still images. Secondary reports quote specific detection percentages from the card; we couldn’t independently confirm those figures, so we’re not repeating them here.

Is it worth trying?

If you already use ChatGPT for images, there is nothing to lose: Images 2.6 is included on every plan, and GIFs and reference images are genuinely new capabilities rather than incremental tweaks. Reference images are probably the most useful day-to-day addition, since character and subject consistency has been a long-standing weak spot for AI image tools.

For developers, the decision is mostly about the speed/quality trade-off. With identical token prices, start with Flare for drafts and high-volume jobs and switch to Sunburst where quality and edit consistency matter. Just budget with real test runs rather than assuming a fixed per-image price, and keep an eye on OpenAI’s docs as quotas and GIF limits are clarified over the coming weeks.