OpenAI's Images 2.6 brings animated GIFs to ChatGPT

OpenAI says users now generate well over four billion images a week through ChatGPT and the API, one billion more than in September. That scale turns the Images 2.6 release into a test of production mechanics, not just a polished cat demo.

The Verge reports that OpenAI's new Images 2.6 model is rolling out to all ChatGPT and Codex users on web, desktop and mobile apps. Its clearest consumer addition creates animated GIFs from text prompts or uploaded photos, then lets users adjust animation speed or edit individual frames.

What can users make now?

OpenAI says either route can produce an animated GIF. After generation, users can change playback speed or edit individual frames rather than accept every image as fixed.

The rollout gives users two starting points:

• A text prompt: The user describes the scene OpenAI should animate.
• An uploaded photo: The user supplies an existing image as the starting point.

Speed control changes how the sequence plays. Frame editing gives the user a way to correct one image. Together, they matter because motion introduces a failure mode absent from a still: the subject can change from frame to frame.

In plain terms, ChatGPT now offers a direct route from words or a photo to an adjustable animation. The useful addition is control after generation, not merely the GIF itself.

How does OpenAI build the GIF?

OpenAI demonstrated the feature with the exact prompt: “Can you make a gif of a cat dancing?” The model produced several slight variations of the character and strung them together flipbook-style into a GIF. Movement comes from sequencing related images, not from one image containing an entire motion sequence.

Images 2.6 has to solve two linked problems here. It must produce separate but related frames, then keep the character recognizable across them. The Verge's comparison with earlier models points to that exact problem: a sequence can contain competent images and still fail because the subject appears to change.

The important technical question is whether those variations stay coherent. The Verge's report says Images 2.6 appears strong enough to keep the cat recognizable, while earlier models would likely have altered its look too much for the trick to work.

That is a qualitative demonstration, not a measured result. OpenAI's example shows the intended effect without quantifying its limits. In plain terms, OpenAI has shown frame-by-frame GIF creation with plausible subject consistency, but has not yet shown how reliable that consistency will be.

What changes for developers?

OpenAI is also releasing two API variants under the GPT-Image-2.6 name. OpenAI says both variants first appeared with Image 2.5, so the new release continues two established tiers.

• GPT-Image-2.6 “Flare” targets efficient image generation.
• GPT-Image-2.6 “Sunburst” targets higher-quality image generation.

The naming also shows continuity in OpenAI's model range. Flare and Sunburst first appeared with Image 2.5, so Images 2.6 updates an established split rather than replacing it with a single model. The source does not say whether API prices, limits or behavior carry over from Image 2.5.

The split gives developers a simple choice between efficiency and quality. Flare targets efficient operation, while Sunburst targets output quality. The announcement, as described in the source, gives no prices, latency numbers, quality scores or rate limits, so developers cannot calculate the trade-off from public figures.

OpenAI also says Images 2.6 should speed up image generation overall, a goal that reaches beyond animated GIFs. The source does not say the API variants expose ChatGPT's GIF controls, so the consumer feature and developer models should not be treated as one documented interface.

In plain terms, the API release offers two performance positions. Their practical difference remains unpublished.

How large is image generation now?

OpenAI's weekly figure covers both ChatGPT and the API. It does not disclose the split between them, the number of people behind the images, average images per user or paid usage. Those omissions limit what the four-billion total can say about actual usage patterns.

Even so, the weekly total is one billion higher than in September, which shows how heavily users now lean on OpenAI's image tools. The practical stakes are speed and consistency at high volume, not merely whether a cat can dance. OpenAI's speed goal targets shorter waits and more efficient generation, while its consistency work targets usable sequences.

The release gives no comparison with competing image generators, so OpenAI's speed claims remain untested against the wider market. At four billion weekly generations, even small improvements can affect how comfortably the company serves image-heavy workflows. In plain terms, image generation is already a scaled workload, so speed now sits beside visual quality as a market issue.

What does the broad rollout mean?

The Verge reports that Images 2.6 is rolling out to all ChatGPT and Codex users across web, desktop and mobile apps. That puts animation in front of OpenAI's broad audience instead of limiting it to API developers or a small preview.

The phrase “rolling out to all users” gives Codex users the same headline capability as ChatGPT users, even though the source does not explain how animated images fit into Codex. The announcement also supplies no start date, plan limits, language support, generation quotas or country availability.

The rollout wording does not say whether access arrives at once for every account. Those gaps leave the practical meaning of “all users” open to interpretation. In plain terms, OpenAI has chosen broad distribution, but the operating details that would determine daily use remain unclear.

How strong is the evidence?

Frame editing gives users recourse when one generated image fails. They can change that frame instead of rebuilding the entire sequence, while speed control changes how the result plays. That makes the workflow more forgiving than a single-shot generator, although the source does not say how often corrections are needed.

OpenAI's demo establishes intended behavior, while the four-billion figure establishes demand. Neither measures how reliably Images 2.6 handles longer sequences, complex scenes or varied subjects. The source offers no frame count, clip length, resolution, generation time, success rate or outside evaluation.

Developers face a similar gap. The names Flare and Sunburst indicate which side of the efficiency-versus-quality trade-off to favor, but they do not quantify the difference. Any comparison remains provisional until OpenAI supplies latency and quality measurements or independent tests supply them.

What should readers watch next?

Three questions will determine whether the release matters beyond GIFs:

• Does Images 2.6 maintain subject consistency beyond OpenAI's cat example?
• Do the claimed speed improvements appear in measured generation times?
• What prices and benchmark results define the gap between Flare and Sunburst?

None of these questions requires a new product category. Each would turn the current directional claims into evidence a developer can use.

The next test is measurable: whether Images 2.6 can hold subject identity across longer clips, speed up real requests and justify the quality gap between Flare and Sunburst. Those results will determine whether the GIF launch marks a durable shift in high-volume image generation or remains a convenient demo.