OpenAI Images 2.6 brings GIFs to ChatGPT and cuts generation time in half
OpenAI rolled out Images 2.6 on November 4, 2026 across ChatGPT, Codex and the API. ChatGPT can create animated GIFs, generation is about 50% faster, and developers get GPT Image 2.6 Flare and Sunburst plus markup, SVG export and a complexity control.
Four billion images a week. That is the scale OpenAI now claims for image generation across ChatGPT, Codex and the API, more than half a billion images on an average day. On November 4, 2026 the company shipped Images 2.6 into that flood: faster stills, native animated GIFs in ChatGPT, and two named API models that trade speed against detail. If you design campaigns, ship product screenshots or build creative tools on top of OpenAI, this release is the first time motion and structured editing land in the same place as the still image stack.
This article sticks to what OpenAI published and what independent coverage verified. Primary source is OpenAI's Images 2.6 announcement. Secondary coverage comes from The Verge, OpenAI's help center and developer docs for Flare and Sunburst, and reporting from heise. After the facts, a short djcroman take section is clearly labelled as opinion.
What shipped on November 4
Images 2.6 is live globally in ChatGPT and Codex on every subscription tier, on desktop, mobile and web. OpenAI says the same generation stack also powers the API through two model IDs: GPT Image 2.6 Flare and GPT Image 2.6 Sunburst. The company frames the release around three promises: speed, motion, and more precise editing rounds that keep characters, products and brand details consistent.
The speed claim is a relative number, not a wall clock for every prompt. OpenAI says Images 2.6 generates images 50% faster than before. In the API, Flare is the fast path: with quality set to low, it can return an image in as little as about six seconds. Sunburst is the detail path: at medium or high quality it is positioned for higher fidelity and stronger consistency across successive edits. Flare also gets credit for better layout quality on posters, product imagery and other professional visuals.
Motion is the consumer headline. ChatGPT can now create animated GIFs from a prompt or from an uploaded still. OpenAI's blog shows sprite sheet style prompts that ask for looping character animations. The Verge adds the interaction detail that matters in the product: after a GIF is made, you can change playback speed, inspect individual frames and share the result. OpenAI's help center matches that: open the GIF to adjust speed, save it or share it; you can also ask for edits in the conversation. One hard limit is clear: ChatGPT cannot edit an animated GIF you upload yourself. GIF creation is also rolling out gradually, so not every account sees it on day one.
The volume number, and how The Verge reads it
OpenAI's wording is careful: people create 4 billion images per week across ChatGPT, Codex and the API. That is not ChatGPT alone. The Verge reports the same 4 billion weekly figure and pairs it with a useful baseline: about 3 billion per week in September 2026. If that September number used a comparable surface mix, weekly volume rose by roughly one third in about two months. OpenAI also translates the weekly figure into a daily average of more than half a billion images.
Those numbers matter for two reasons. First, they show that generative images are no longer a niche side feature inside ChatGPT; they are a primary workload. Second, they explain why OpenAI is pushing latency and editing controls so hard. When hundreds of millions of images are requested every day, shaving generation time and reducing failed edit rounds is an infrastructure problem as much as a creative one.
Treat the absolute totals as company reported. OpenAI did not publish a methodology breakdown by product or by free versus paid users. The Verge's September comparison is useful context, not an independent census.
What creators can do with GIFs
OpenAI positions GIFs as the motion layer for social, marketing and early storyboarding. Examples on the blog include animating a brand mascot, building reaction clips for social feeds, and exploring character movement before a full video shoot. A quote from OpenArt CEO Coco Mao frames GIFs as the native language of social media and says GPT Image 2.6 lets creators speak that language from a single prompt. Town, a startup building an AI work assistant, is cited as an early API customer exploring animated characters for everyday workplace interactions.
The help center makes the workflow concrete. Ask for a looping GIF and describe what should move, or upload a still and ask ChatGPT to animate it. Afterward, change the motion with another text prompt, then open the result to tweak speed. Saved reference images are another related upgrade: you can name photos and reuse them later without uploading again. That is not exclusive to GIFs, but it helps multi turn character work that GIFs will amplify.
What this is not: a replacement for dedicated video models or professional motion graphics. GIFs are short loops. OpenAI still points people who need longer video elsewhere. The honest use case is faster drafts for social, pitches and early character tests, not final broadcast packages.
What developers get in the API
Images 2.6 is not only a ChatGPT skin. Developers get three control surfaces that used to be awkward or missing in the API.
First, image markup comes to the API after already shipping in ChatGPT. Teams can mark the exact region that should change instead of rewriting a full prompt and hoping the model finds the right corner of the frame. That is the difference between "make the logo bigger" and pointing at the logo.
Second, SVG export lands for designs that need to travel into websites, apps and print. Vector output matters when you want clean edges at any size. Heise notes an important cost detail: SVG work can cost more because OpenAI first generates a reference image and then converts it with another model. Budget for that second pass if your pipeline is heavy on icons or logos.
Third, a new complexity parameter lets developers dial how much planning the model does. Quality still controls visual detail and fidelity. Complexity guides how much structure goes into the scene. OpenAI's guidance is practical: for most edits, use low complexity for speed and pair it with the quality level your workflow needs; for scenes with many elements or hard constraints, try medium or high complexity.
Figma Weave creative director Itay Schiff is quoted saying GPT Image 2.6 brings speed and creative control together and that the team is adding it to Figma Weave for creative production inside a reusable workflow. That quote is marketing, but it signals where OpenAI wants the API to sit: inside design tools, not only in one off playground prompts.
Pricing for Flare and Sunburst, per OpenAI's developer docs, is the same table for both models: $5 per million text input tokens, $8 per million image input tokens, $30 per million image output tokens, with cheaper cached input rates. Heise reports those prices are unchanged from Images 2.5. If you already modelled costs on 2.5, the bill structure does not jump with the 2.6 names; you are paying for a faster and more controllable version of the same token economy.
Limits OpenAI still admits
OpenAI's own materials still flag familiar weak spots: text inside images, complex layouts and character consistency. Images 2.6 is marketed as better on consistency across edits, especially with Sunburst at medium or high quality, but the company does not claim the problems are solved. Provenance tooling continues: C2PA metadata and invisible SynthID watermarks remain part of the stack, according to secondary reporting.
Availability has caveats too. The ChatGPT and Codex rollouts are described as global across tiers, yet GIF creation is explicitly gradual. If your workspace does not see GIF controls yet, that matches the help center, not a broken account. Uploaded animated GIFs still cannot be re edited as animations. And six second Flare generations are a low quality setting, not a promise for every high detail marketing render.
How this fits the 2026 image race
By late 2026, almost every major assistant can make a still. The fight has moved to latency, edit control, brand consistency and whether motion is native or bolted on. Google, Midjourney adjacent tools and open weight image stacks all compete on some of those axes. OpenAI's answer with Images 2.6 is to keep the volume advantage, cut wait time in half again, put GIFs inside the chat surface hundreds of millions already use, and give API customers markup, SVG and complexity knobs without changing the headline token price.
The Verge's September to November jump, from about 3 billion to 4 billion weekly images, is the demand signal behind that product bet. When a feature is already generating billions of assets a week, shipping motion and faster edits is less about a demo and more about holding the default slot in people's creative workflow.
djcroman's take: motion inside the chat wins the next round of casual creation
This section is opinion. The paragraphs above are reported fact; what follows is how djcroman reads the release.
The important product move is not "OpenAI can make GIFs." Plenty of tools can. The important move is that motion now lives in the same chat thread where people already brief stills, iterate on brand colors and export a draft to Slack. That collapses a two tool workflow (chat for the idea, CapCut or a GIF maker for the loop) into one. For marketers and social teams, that is worth more than another tiny jump on a static image leaderboard.
The API split into Flare and Sunburst is the grown up version of the same idea. Creative tools need a fast preview model and a slower hero model. Naming them, documenting complexity, and keeping prices flat from 2.5 makes it easier for product teams to wire both into one pipeline: Flare for the brainstorm carousel, Sunburst for the final pack shot that must survive five markup rounds.
There are real limits to the hype. GIFs are not video. Six seconds at low quality is not a campaign final. Character consistency is still an OpenAI admitted weak spot. And company reported volume figures should not be treated like audited traffic. Even so, the direction is clear: the winners in consumer generative imagery will be the products that remove switches between tools, not the ones that win a single still frame beauty contest.
What you can do now
- In ChatGPT, try a looping GIF prompt or upload a still and ask for animation, then open the result to change playback speed.
- If GIF controls are missing, wait for the gradual rollout rather than assuming a billing issue.
- In the API, route interactive previews to Flare at low quality and finals to Sunburst at medium or high quality.
- Prefer low complexity for simple edits; raise complexity when the scene has many constraints.
- Budget extra for SVG jobs if your pipeline converts many assets to vectors.
- Keep provenance expectations realistic: C2PA and SynthID help platforms, they do not stop every misuse.
Sources
Source: openai.com