Four billion images a week. That’s the number OpenAI is putting behind Images 2.6, and it explains why this release focuses on speed and repeat editing instead of a new look. When people generate that many images, the bottleneck is no longer the first result. It’s the fifth revision.
OpenAI announced the model for ChatGPT, Codex, and the API. The pitch has three parts: images generate 50% faster, characters and products stay consistent across edits, and GIFs are now a supported output.
Two API models, one speed-quality split
Developers get two options: GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst. Flare is the fast one. At low quality it can produce an image in as little as six seconds, and OpenAI says it also improves layout for posters and product imagery. Sunburst is for detail and edit-to-edit consistency at medium or high quality.
This is a familiar split. Google’s Imagen line and Black Forest Labs’ FLUX family have long offered fast and slow tiers, and fal and Replicate users are used to choosing between them. What’s new is that OpenAI now does it inside one product family, which makes routing simpler.
GIFs are the splashy part
The animation feature works by having the model produce sprite sheets, which become looping GIFs. OpenAI’s own example prompt asks for exactly six frames in a 3-by-2 grid. That’s a clever workaround, not true video generation. It won’t replace Sora, Veo, or Runway for serious motion work. But for reaction memes, mascots, and animated storyboards, it’s a cheap and fast route.
OpenArt’s CEO called GIFs the native language of social media, which is the obvious marketing angle. Town, a startup building a work assistant, is already using it for an animated typing-indicator character. That’s a small use case, but a realistic one.
What developers actually get
- Image markup in the API, so teams can point to the exact area to change
- SVG export for use across web, apps, and print
- A complexity parameter that sets how much planning the model does
- Separate quality controls for visual detail and fidelity
The complexity setting is the most interesting item here. Quality and complexity are now separate dials. OpenAI recommends low complexity for most edits and medium or high for scenes with many elements or constraints. That should cut cost and latency for routine work, though OpenAI hasn’t published numbers in the announcement.
SVG export is also a real gap-filler. Designers have complained for years that raster-only output makes AI images awkward to move into production. Adobe Firefly and Recraft already offer vector output, so OpenAI is catching up, not leading.
Pricing and availability
Images 2.6 is live globally in ChatGPT and Codex across all subscription tiers, on desktop, mobile, and web. Both Flare and Sunburst are in the API, with pricing on OpenAI’s site. A system card covers safety evaluations.
Does it matter?
Partly. Early partners including Figma Weave, Picsart, fal, and Higgsfield are already building on it, which signals distribution matters more than any single demo. The GIF feature is fun but narrow. The real story is iteration: faster drafts, tighter consistency, and finer control over how much compute each edit burns. If Sunburst holds character and brand details across many rounds, that’s what teams will pay for. Test it on your own edit chains before believing the claims.
