OpenAI Images 2.6: ChatGPT Gets GIFs — and the API Price Behind Them

ChatGPT can make GIFs now. OpenAI announced Images 2.6 on November 4, 2026, and the headline feature is exactly what it sounds like: describe a scene in a sentence or upload a photo, then get back an animated loop you can speed up, slow down, and edit frame by frame. No timeline software, no sprite-sheet wrangling. It is the first time native GIF creation has shipped inside ChatGPT, and it is rolling out to every tier, free included.

There is more under the hood than the meme maker. Images 2.6 arrives with a 50% faster generation claim, reusable reference images that keep your characters and pets consistent across prompts, and two new API models: GPT-Image-2.6 Flare for speed and GPT-Image-2.6 Sunburst for quality. OpenAI also pushed image markup into the API, added SVG exports, and introduced a complexity control that decides how much planning the model does per request.

Here is the full breakdown: what launched, how the two API models differ, what the rate card actually says (spoiler: Flare and Sunburst cost exactly the same), and the fine print worth reading before you build anything on it.

What Images 2.6 adds to ChatGPT

The GIF feature is the part you will notice first. In ChatGPT, you can now turn a written prompt or an uploaded photo into a short animated loop in a single step, then adjust its playback speed and edit individual frames before sharing. OpenAI pitches it at marketers animating a brand mascot, creators making reactions and memes, and designers exploring character movement before committing to video production.

Early access partners shaped the announcement. OpenArt, which called GIFs "the native language of social media," has early access creators building on it; Town, a startup making an AI work assistant, used the API version to explore animated characters for workplace interactions; and Figma is adding GPT-Image-2.6 to Figma Weave for creative production workflows. Picsart, fal, Sable, and Higgsfield were also named as early access customers. The partner quotes matter because image-generation launches are notoriously hard to judge from company demos: real creative teams stressing the frame editing is more informative than a chibi office worker sprite sheet.

Equally important for regular users: reusable reference images. You can save images of yourself, your pets, or your characters and reuse them across new generations without uploading again. This is the consistency problem every AI image tool fights (your character drifts between generations), and persistent references are the practical fix. OpenAI announced this via its launch posts alongside the GIF rollout, and it lands in ChatGPT and Codex the same day.

Speed is the third pillar. OpenAI says Images 2.6 generates images 50% faster than before, and gives a concrete API figure: at low quality, Flare produces an image in as little as six seconds. The company also reported that weekly image volume across ChatGPT, Codex, and the API has reached 4 billion images, up from 3 billion in September, when Images 2.5 launched. That scale explains the urgency: at more than half a billion images a day, every second shaved off generation time compounds fast.

Flare vs Sunburst: the two API models

Developers get two models, continuing the two-variant strategy OpenAI used for Images 2.5. Here is how the company positions them:

The important detail: they are priced identically. Flare is not the cheap one and Sunburst is not the premium one. The choice is latency versus fidelity, not money. Sunburst also improves image consistency across edits when quality is set to medium or high, which is the setting that matters most for teams iterating on brand assets. Flare, for its part, gets better layout quality for posters and product imagery than its 2.5-generation predecessor.

What the rate card actually says

OpenAI's developer pricing page now lists both gpt-image-2.6-flare and gpt-image-2.6-sunburst. Here is the official rate card, per 1 million tokens, with November 5, 2026 market-rate approximations for UK and Canadian readers (1 USD = 0.7415 GBP, 1.3974 CAD, verified at publication):

Three things stand out. First, output tokens are the expensive part: $30 per million versus $8 for image input. Whatever you generate, the pixels you receive cost nearly four times the images and text you sent. Second, the cached input rows ($2.00 image, $1.25 text) are the big lever for editing workflows: revise the same image repeatedly and your repeated input tokens bill at a quarter of the standard rate. Third, the rates apply to both Flare and Sunburst identically, confirmed row by row on the live pricing page.

What you cannot do from this table alone: quote a price per image. OpenAI bills in tokens, not images, and the token cost of one image depends on quality tier and size. The company's image-generation guide includes an input cost calculator for this: if you are budgeting a production run, use that instead of guessing from these per-million figures. Any coverage that quotes a neat "this many cents per image" without citing OpenAI's calculator is doing arithmetic you should not trust.

The "50% faster" claim, read carefully

OpenAI says Images 2.6 generates images 50% faster than before. What does that actually mean? The one concrete benchmark published alongside the launch is OpenAI's own Flare figure: generation time falling from 7.08 seconds with Images 2.5 to 4.84 seconds with Images 2.6. That is roughly a 32% reduction in elapsed time, or about 46% more images produced in the same period.

Both framings can be true at once: 50% faster can describe throughput (images per hour), while the stopwatch shows a 32% drop in seconds per image. AlphaSignal's coverage did the math and flagged the distinction. The honest version: OpenAI's own benchmark shows a genuine, measured improvement in the same conditions (low quality, Flare variant), but it is not a 50% reduction in the time you wait. Do not read it as "everything renders twice as fast."

No independent head-to-head exists yet. The 7.08s→4.84s figures are OpenAI's own measurements on its own model, and no third party has published replicated latency tests. That is normal for a day-one launch; it is not a reason to dismiss the claim, but it is a reason to file it under "vendor-reported" until someone else times it.

New API controls: markup, SVG, and complexity

Beyond the two models, the API gets three workflow features worth more attention than the GIF headline:

• Image markup comes to the API. ChatGPT already lets you mark up an image to point at what to change; now developers can build the same "edit exactly this region" behavior into their apps. For design tools and e-commerce teams iterating on product shots, region-targeted edits are more valuable than raw generation speed.
• SVG exports. Designs can now export as vector graphics, ready for websites, apps, and print. This is the bridge feature for anyone producing logos, icons, and illustrations; raster-to-vector workflows have always been a pain point, and native SVG output removes a conversion step.
• A complexity parameter. Developers can dial how much planning the model does per request. Lower complexity suits quick edits; medium or high is for scenes with many elements or constraints. Paired with the existing quality control, this gives a speed-vs-fidelity slider that maps neatly onto the Flare/Sunburst split.

Taken together, these say the release is really aimed at production design workflows (teams embedding image generation into products) more than at the casual GIF user. The casual user is the marketing; the API controls are the product strategy.

Rollout and availability: what is confirmed

OpenAI says Images 2.6 is available globally in ChatGPT and Codex across all subscription tiers on desktop, mobile, and web. That includes free users: GIF generation carries no separate paywall, though your account's normal usage caps still apply. The rollout began on November 4, 2026, and reporting (The Verge first, picked up widely) describes it as a gradual push rather than a simultaneous switch-on. If you do not see the GIF option yet, that is the rollout reaching you, not a tier restriction.

A word of caution on the API side: some coverage noted that OpenAI's API guides still referenced the 2.5 generation as of launch day. The official pricing page now lists the 2.6 models, and the announcement confirms Flare and Sunburst are live in the API, but if your SDK or docs cache shows older model names, check the current pricing page rather than assuming.

Common mistakes and catches before you build on this

Do not budget per image. The rate card is per million tokens, and per-image token costs vary by quality tier and size. The most common pricing error with image APIs is multiplying a per-image guess from one tool by volume from another. Use OpenAI's own cost calculator on the image-generation guide for your exact configuration.

Do not pick Flare to save money. Both models share the identical rate card. Flare saves time, not dollars; at low quality that is the point. If you are generating thousands of assets, the lever that matters is the cached input row, not the model name.

Watch the regional uplift. OpenAI's pricing page states that regional processing (data residency) endpoints carry a 10% uplift for models released on or after March 5, 2026. Images 2.6 is well past that cutoff, so EU or UK data-residency deployments will bill 10% above the standard rates. FedRAMP endpoints carry the same uplift.

Treat the 50% figure as throughput, not stopwatch. The measured number is 7.08 to 4.84 seconds (~32% less waiting) at low quality on Flare. Medium and high quality on Sunburst will be slower (OpenAI does not publish those times). Quote "50% faster" as the company's claim, not as a measured universal fact.

Reference images are the sleeper feature. For anyone generating character or product series, reusable references will matter more than GIFs over time. Test how well your references survive multiple edit rounds before committing a pipeline to them: consistency is the claim every vendor makes and the one that degrades fastest in practice.

FAQ

Can I make GIFs in ChatGPT for free?

Yes, according to OpenAI's rollout details. Images 2.6 is available across all ChatGPT and Codex tiers, and GIF generation has no separate paywall. Your account's normal usage caps still apply, so heavy use will hit the same limits as regular image generation.

What is the difference between GPT-Image-2.6 Flare and Sunburst?

Flare is optimized for speed, with as little as six seconds per image at low quality. Sunburst is optimized for detail and edit consistency at medium or high quality. They cost exactly the same per token, so the choice is about latency versus fidelity, not budget.

How much does the Images 2.6 API cost?

OpenAI's published rate card is $8 per 1M image input tokens and $30 per 1M image output tokens, with cached input at $2 and text input at $5 per 1M. Those are token prices, not per-image prices: your cost per image depends on quality tier and size, so use OpenAI's cost calculator for exact numbers.

Can the API output GIFs directly?

OpenAI's announcement presents GIF creation as a ChatGPT feature, and the API announcement focuses on Flare and Sunburst, image markup, SVG exports, and the complexity parameter. The company's launch materials do not confirm direct GIF output as an API format, so treat that as unconfirmed for now.

Is Images 2.6 available in my country?

OpenAI says the rollout is global across ChatGPT and Codex on web, desktop, and mobile. It is a gradual rollout that began November 4, 2026, so the GIF option may appear at different times for different accounts. Data-residency deployments in the EU/UK carry a 10% pricing uplift.

Is Images 2.6 actually better than Gemini or Grok's image models?

Nobody can say yet. OpenAI published no head-to-head benchmarks against rival image models, only its own 2.5-to-2.6 latency comparison. Independent testing typically follows a launch by days or weeks; treat any "best image model" claims as marketing until those results exist.

Sources

• OpenAI, "Introducing Images 2.6," November 4, 2026 — openai.com (official announcement; GIF features, Flare/Sunburst, 4B images/week, rollout details)
• OpenAI Developer Pricing — developers.openai.com (official rate card; gpt-image-2.6-flare/sunburst rows, cached input, regional uplift)
• Unite.AI, "ChatGPT Adds Animated GIF Creation and Editing with Images 2.6" — unite.ai (The Verge's Nov 4 report, all-tier availability, 2.5 comparison)
• AlphaSignal, "OpenAI's Images 2.6 Generates 4B Weekly Images 32% Faster With Native GIFs" — alphasignal.ai (7.08s→4.84s benchmark math, no-paywall detail, model positioning)
• TestingCatalog, "OpenAI launches Images 2.6 with GIFs and image reference" — testingcatalog.com (reference images, OpenAI X-post quotes, API pricing page confirmation)
• Crypto Briefing, "OpenAI rolls out Images 2.6 globally across ChatGPT and Codex" — cryptobriefing.com (gradual rollout, all-tier availability)
• Runtimewire, "OpenAI launches ChatGPT Images 2.6 with GIFs, reference images and faster API models" — runtimewire.com (Flare 6-second claim, partner context)
• Xenospectrum, "ChatGPT Can Now Generate Animated GIFs From Text or Photos" — xenospectrum.com (demo-vs-verified caution, API-format caution)

Internal reading: our Sora alternatives guide, FLUX 3 Max review, Ideogram Vector 1.0 review, and Gemini Omni Flash review.