OpenAI Images 2.6 brings animated GIFs, SVG export, and two new API models to all tiers
OpenAI's Images 2.6 adds animated GIFs, reusable references, SVG export, and area markup, plus two API models: Flare (speed) and Sunburst (quality).
OpenAI Images 2.6 brings animated GIFs, SVG export, and two new API models to all tiers
Just two months after releasing Images 2.5, OpenAI has shipped Images 2.6 across ChatGPT, Codex, and the API. The update lands globally for all subscription tiers, free through enterprise, and brings a meaningful set of new capabilities rather than a quiet incremental bump.
The headline numbers give a sense of the scale this sits within: users are now creating 4 billion images per week across ChatGPT, Codex, and the API. That is more than half a billion per day. OpenAI is clearly treating image generation as a core product surface, not a side feature.
What’s new in Images 2.6
Animated GIF output. You can now describe a scene and the motion you want, or upload a photo, and ChatGPT will produce an animated GIF. The feature also works with rough sketches: the model first develops the sketch into a complete image, then adds animation. Once generated, you can adjust playback speed and inspect individual frames to refine specific areas. One limitation worth noting: ChatGPT cannot edit an animated GIF you upload yourself. This feature only applies to GIFs it generates.
Reusable reference images. You can store reference images of yourself, a pet, a character, or a product, and call on them again in future conversations without re-uploading. This is a practical quality-of-life change for anyone doing consistent creative work across sessions.
SVG export. Generated images can now be exported as SVG files, making them suitable for use on websites, in apps, and in print without quality loss at any scale. The technical note here matters: SVG output goes through the standard generations endpoint with output_format: "svg" and n set to 1, but the model generates SVG from a prompt rather than vectorising an existing image. If you want to convert your existing logo or artwork to SVG, this is not the tool for that.
Area-specific markup in the API. The markup tool, which lets you highlight a specific region of an image for targeted edits, was already available in ChatGPT. With Images 2.6 it comes to the API, giving developer teams the same precision editing capability in their own workflows.
50% speed improvement. Image generation is 50% faster than Images 2.5, which was itself 50% faster than Images 2.0. The cumulative gain matters for anyone iterating quickly.
The two new API models: Flare and Sunburst
For developers accessing image generation through the API, OpenAI has introduced two distinct models rather than a single update.
GPT-Image-2.6 Flare (gpt-image-2.6-flare) is optimised for speed and interactive workflows. Set quality to low and it can produce an image in as little as six seconds. That is useful for prototyping, rapid iteration, or any application where responsiveness matters more than maximum detail.
GPT-Image-2.6 Sunburst (gpt-image-2.6-sunburst) is optimised for quality and consistency. Set quality to medium or high and you get more detailed compositions with stronger coherence across edits. OpenAI notes that Sunburst took first place on the Design Arena benchmark, displacing its own predecessor. Both models support complexity control, markup-guided edits, SVG output, and animated GIF output.
Pricing is identical for both models and matches what OpenAI charged for GPT-Image-2.5:
| Modality | Input | Output | Cached input |
|---|---|---|---|
| Image | $8.00 / 1M tokens | $30.00 / 1M tokens | $2.00 / 1M tokens |
| Text | $5.00 / 1M tokens | , | $1.25 / 1M tokens |
The choice between Flare and Sunburst is therefore entirely about what your use case demands, not about cost.
What this means for you
If you use ChatGPT for creative work, the GIF output and reusable reference images are the features most likely to change your day-to-day workflow. Creating short animated loops from characters, product shots, or sketches is now a prompt away, and not having to re-upload reference material every session is a small but genuine convenience.
If you build on the OpenAI API, the split into Flare and Sunburst gives you a cleaner decision to make: reach for Flare when speed is the priority, Sunburst when output quality and edit consistency matter. The arrival of markup in the API means you can now build precision editing flows that previously required the ChatGPT interface.
If you are evaluating OpenAI’s image generation against alternatives, the two-tier API model structure is worth watching. Having a named speed model and a named quality model makes it easier to reason about trade-offs in product design, and the pricing parity between them removes cost as a complicating variable.
Safety and provenance
OpenAI has built on the same safety foundations used for Images 2.5, with additional safeguards specific to GIF creation. All images generated through ChatGPT, Codex, and the API are watermarked using Google DeepMind’s SynthID for provenance tracking.
Images 2.6 is available globally from November 2026 across all ChatGPT tiers on desktop, mobile, and web, as well as in Codex. The API models are available now through the OpenAI API.