ChatGPT Can Now Make Animated GIFs, and OpenAI Split Its Image Model in Two

GIFs come to ChatGPT

OpenAI launched ChatGPT Images 2.6 on November 4, and the headline feature is one users have wanted for years: ChatGPT can now make an animated GIF. Describe a short clip, or upload a photo, and ChatGPT returns a looping animation instead of a still image. You can adjust the playback speed and edit individual frames inside the chat before sharing.

The update is rolling out globally in ChatGPT and Codex on web, desktop, and mobile, across all subscription tiers including free. The GIF feature itself is rolling out gradually, so it may not appear in your account immediately. OpenAI also notes that you cannot upload an existing GIF and edit it. The feature only makes new ones.

The numbers behind the launch are large. OpenAI says users now generate more than 4 billion images a week across ChatGPT, Codex, and the API, up from 3 billion in September. Image generation has become one of the most-used features in ChatGPT, and the GIF support is OpenAI's bet that short loops are the next format people want.

Two models, two jobs

For developers, the release splits the image model in two. The API now offers GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst, continuing the two-model strategy from Images 2.5.

Flare is the fast model. OpenAI says it can return a low-quality image in as little as six seconds, and it is tuned for interactive work: drafts, posters, product variations, and quick edit rounds where you will throw away half the outputs. Sunburst is the quality model, tuned for detail and consistency across edits, for brand assets, product shots, and anything with many constraints.

The practical rule is simple. Use Flare while you are exploring, and switch to Sunburst when the design is fixed and you need the final version. If you are building image generation into an app, set the model explicitly in your API calls, because ChatGPT itself does not expose a variant selector.

The 50 percent faster claim deserves a closer look

OpenAI says image generation is 50 percent faster than Images 2.5. That figure needs context. The only absolute latency number in the announcement is Flare at low quality: as little as six seconds. An independent read of OpenAI's own Flare benchmark puts the improvement at about 32 percent, from 7.08 seconds to 4.84 seconds, which is still a real gain, but not the one in the headline.

Treat the speed claim the way you should treat any vendor benchmark: as a direction, not a measurement. If speed matters to your workflow, time your own prompts before and after the switch. The improvement is real, but your mileage will depend on image size, quality settings, and how busy the service is.

What else shipped

Beyond GIFs and the two models, Images 2.6 adds several features that matter more to professionals than the GIF button does.

• Animated GIF output in the API, with frame counts from 2 to 64 and per-frame duration control. The API returns a finished animation, not a sprite sheet you assemble yourself.
• SVG export for developers, so generated graphics can scale cleanly.
• Image markup in the API, for annotating and editing specific parts of an image.
• A complexity setting that controls planning separately from image quality.
• Reusable reference images, so a character or brand detail stays consistent across generations.

OpenAI cites early partners using these features for sprite-sheet GIFs, animated assistant characters, character consistency, and reusable creative workflows. The message is clear: OpenAI wants image generation to be a production tool, not just a toy.

A word of caution

Animated images of real people can mislead, and a convincing GIF spreads faster than a still image. OpenAI ran both models on 3,077 adversarial prompts and reports that the share of unsafe generations not blocked fell slightly, from 1.92 percent to 1.72 percent for Sunburst. Those are test numbers, not production rates, and they are a reminder that the safety systems are probabilistic.

In the EU, the AI Act's transparency rules have applied in full since August 2, 2026, which means synthetic content must be marked in a machine-readable format. If you publish AI GIFs, label them. It is the law in Europe, and it is good practice everywhere else.

If you are sorting out which AI tools are worth paying for, our AI tool stack guide will help you cut the list down to six tools that actually earn their keep. And if you want to get better at telling real video from generated video, read how to spot an AI-generated video before you share the next clip that lands in your feed.