OpenAI GPT-Image-2.6 Makes GIFs and SVGs, and the 50% Faster Claim Needs a Stopwatch
OpenAI released GPT-Image-2.6 on November 4 with animated GIFs, SVG export, and two API models called Flare and Sunburst. The speed claim is untested, but the new controls are real.

GPT-Image-2.6 adds GIFs, SVG export, and two new API models
OpenAI released GPT-Image-2.6 on November 4, 2026, and it is the most practical image model update the company has shipped in a while. The headline features are animated GIF output, SVG vector export, image markup in the API, and a new complexity setting that controls how much the model plans a scene before drawing it. GPT-Image-2.6 is live in ChatGPT, in Codex, and through the API on every ChatGPT tier.
The API ships it as two models with different jobs. GPT-Image-2.6 Flare is tuned for speed, and OpenAI says it can return an image in as little as six seconds at low quality. GPT-Image-2.6 Sunburst is tuned for detail and for holding a subject steady across a chain of edits. Both cost the same per token, which makes the choice about workflow rather than budget.
OpenAI also says the service now generates more than 4 billion images a week across ChatGPT and the API, up from 3 billion in September. That volume is why small workflow changes in this release matter more than they look.
Flare is fast, Sunburst is careful
The split between the two models is the most useful part of the release. Flare is the one to reach for when you are iterating: posters, product shots, layout drafts where text and objects need to sit in the right place. OpenAI says Flare also improves layout quality for those composition-heavy cases.
Sunburst is the one for multi-round editing, where the usual failure is drift. Change the background in round one, fix a logo in round two, and by round three the product looks like a different product. OpenAI says Sunburst holds people, products, and brand details steadier across edits. That is a claim worth testing on your own brand assets before trusting it in production.
Pricing is unchanged from Images 2.5: $30 per million output tokens, $8 per million input image tokens, $2 cached. One catch: a low-quality image now uses about 4.2 times the output tokens it did on 2.5, so the speed gain is not a price cut in disguise. Batch processing at half price does not cover Images 2.6 yet either.
The GIF feature is the one regular users will notice
In ChatGPT, you can now turn a prompt or a photo into a looping GIF, set the playback speed, step through individual frames, and fix one before sharing. On the API, GIF is an output format alongside PNG, JPEG, and WebP, with 2 to 64 frames.
The obvious uses are brand mascots, reaction loops, and quick storyboards before video production. A mascot that waves is a small thing, but it is the kind of small thing social teams actually ship every day. The GIF rollout in ChatGPT is happening in batches, so if you do not see it yet, that is the rollout, not a missing feature.
There are guardrails worth knowing about. OpenAI tested GIF safeguards against an adversarial set of images and says they catch roughly three in four GIFs already labeled violative. That is not a perfect filter, and anyone publishing GIFs at scale should still review them.
SVG export is the quiet win for developers
SVG export is the feature developers should care about most. It produces vector output straight from the image call, which matters for logos, icon sets, and anything headed for a print or design pipeline that would otherwise re-trace a raster image by hand.
The constraints are tight, and they are worth reading before you build on it. The output is static vector content with no scripts, external resources, or embedded rasters. The number of images per request must be 1, there is no streaming, and neither output compression nor the prompt rewriter works with it. The edits endpoint cannot output SVG, and an SVG cannot be used as a raster input for edits.
Independent testers found SVG requests failing upstream on launch day, then working the next day at a flat $0.0315 per SVG on one third-party platform. Treat that as early-days roughness rather than a broken feature, but do not build a launch-week demo that depends on it.
The 50% faster claim has no public baseline
OpenAI says GPT-Image-2.6 generates images 50% faster than before, but the announcement never says faster than what, and no independent benchmark had listed either model days after launch. Artificial Analysis had neither Flare nor Sunburst on its text-to-image or image-editing boards as of November 5.
That does not make the claim false. It makes it unverified. The six-second figure is a best case at the lowest quality setting, not a typical time, and third-party timing on launch day put Flare at 16 seconds and Sunburst at 28 seconds for the same 1K poster. Your own prompts will land somewhere in that range.
The honest way to read this release: the speed is a nice bonus, the controls are the real upgrade. A complexity setting, image markup, and SVG export change how you write an image call. A faster render just changes how long you wait for it.
What it costs, and who should care
If you generate images in ChatGPT, try the GIF feature and see whether your workflow actually wants loops. If you build with the API, start with Flare for drafts and Sunburst for anything that goes through multiple edit rounds, and watch the token math on low-quality images.
The bigger picture is that image models are becoming design tools rather than slot machines. Google's video-capable Gemini model is pushing the same direction from the video side, and the AI tool stack most teams actually use keeps getting one more layer. GPT-Image-2.6 is a solid step in that direction, even if the speed number deserves a stopwatch rather than a headline.


