GPT Image 2.6: animated GIFs, saved references and 50% faster image generation in ChatGPT

Updated on November 10, 2026

GPT Image 2.6 is OpenAI's image generation model deployed on November 4, 2026 inside ChatGPT, Codex and the API, available as two variants: Flare for speed and Sunburst for detail. According to OpenAI, the platform now produces 4 billion images per week. The update introduces animated GIF output, saved reference images, SVG export, markup-guided edits and a claimed 50% speed improvement. Every ChatGPT plan gets access, free tier included; API billing starts at $30 per million image output tokens.

GIFs, references and markup: what GPT Image 2.6 brings to the table

GPT Image 2.6 produces a looping GIF straight from the chat window. Type a prompt, upload a sketch or drop in a product photo and ChatGPT returns an animation. Playback speed is adjustable after the fact, and each frame can be inspected or retouched individually. OpenAI publishes no duration, frame count or resolution for these GIFs, and the rollout is gradual.

Saved reference images solve the re-upload headache. Store a character, a pet or a product shot under a name, then call it back with an @mention in any later conversation. Sunburst in particular keeps facial features, outfits and brand details more consistent across successive edits.

On the API side, markup editing lets developers mark the exact zone to modify directly on the image instead of describing it in words. The complexity parameter (low, medium, high) controls scene richness without changing per-token cost. SVG export outputs a lightweight vector file ready for web or print.

• GIF output: looping animation generated through the Image API, playback speed adjustable in ChatGPT
• SVG: vector file ready for websites, apps and print
• Markup: zones to edit marked directly on the image, in ChatGPT and the API

Flare vs Sunburst: same price, different strengths

Both models share the same API pricing and are equally available across every ChatGPT plan. The split is practical. Flare targets rapid iteration: thumbnails, campaign variants, A/B testing. Sunburst handles the polished final: detailed product shots, posters with precise text, and multi-turn character consistency.

According to OpenAI, Flare can return an image in as little as six seconds at low quality, and Images 2.6 generates about 50% faster overall than before. Sunburst runs slower, but setting complexity to low, which OpenAI recommends for most edits, reduces latency at no extra fee. One thing to know: per-token pricing is identical to ChatGPT Images 2.5, and no per-image price is published; OpenAI points to a cost calculator instead.

Market position and getting started

In the text-to-image model landscape, GPT Image 2.6 competes with Nano Banana 2.1, Midjourney v8.2, Ideogram 4.5, Seedream 5.0, FLUX 3 Image and Recraft V4.1 in a market where gaps remain tight and shift fast. No independent leaderboard result has been verified for this release; its claimed strengths are native GIFs, saved references and the speed gains OpenAI reports.

Getting started in ChatGPT takes a single sentence describing the image or a photo upload. Developers call gpt-image-2.6-flare or gpt-image-2.6-sunburst through the Image API or Responses API, with documented Python and Ruby SDKs. API rates are listed on OpenAI's pricing page and may change; always check the official source for the latest figures.

Frequently asked questions