GPT Image 2.6

GPT Image 2.6 is OpenAI's image model in ChatGPT and the API that adds looping GIFs, faster generation and markup edits, for creators and developers.

The headline change: a prompt can now return a looping GIF

OpenAI announced Images 2.6 on November 4, 2026. Before it, ChatGPT could not make a GIF natively in one step. Now you ask for a looping GIF and describe what should move, or upload a still image and ask ChatGPT to animate it. The ChatGPT help page says GIF creation is still rolling out, so it may not be on your account yet.

• In ChatGPT you can change a GIF it made by describing the edit in the conversation, change its playback speed, save it and share it.
• ChatGPT cannot edit an animated GIF that you upload.
• In the API, set output_format to gif on the generations or edits endpoint. Use the edits endpoint when still reference images should guide the animation.
• OpenAI's API example is a monkey lifting a lantern: gpt-image-2.6-sunburst, 512x512, 9 frames at 200 ms, a 1.8 second loop that returns to the starting pose. Its tip for a smooth loop is to give the subject a starting pose to return to and say whether the camera moves.

Two API models: Flare for speed, Sunburst for detail

• gpt-image-2.6-flare is the smaller model for workflows where response time matters most. OpenAI says it can generate in as little as six seconds at low quality, and that it improves layout for posters and product images.
• gpt-image-2.6-sunburst is the base model for demanding detail, layout or edit accuracy, with more consistency across edits at medium or high quality.
• OpenAI suggests starting with Flare if you used GPT Image 2 or 2.5 Flare, and Sunburst if you used 2.5 Sunburst or GPT Image 2 fell short, then comparing both on your own prompts.
• ChatGPT and Codex use the same release, on desktop, mobile and web in every subscription tier. OpenAI says images are generated 50% faster than before. That is its own claim, not something Seailife timed.

Other new controls in the API

• complexity (low, medium or high) sets how much the model plans before rendering. It defaults to high and adds no separate fee. OpenAI recommends low for quick edits and warns that higher complexity can add latency and unwanted changes.
• Markup edits on /v1/images/edits take a transparent PNG overlay that marks where a change should happen, with the change described in the prompt. The first image and the overlay can each be at most 16,777,216 pixels, the overlay file at most 50 MB, and markup cannot be combined with a mask.
• SVG output comes from the generations endpoint only, with n set to 1. The edits endpoint does not return SVG and does not accept an SVG as an input image.

GIF limits to check before you build on it

• frame_count runs from 2 to 16, default 9. frame_duration_ms runs from 20 to 2000, default 200, rounded to 10 ms.
• size is auto or width and height in multiples of 16. Frame count, frame size and the number of animations must fit the model's size limits and the request's total pixel budget.
• GIF output has no streaming, no partial images and no custom output_compression.
• The API reference says GIF output needs access to the GIF preview for the chosen model, so check your account before planning a product around it.

What it costs through the API

The pricing page lists the same standard rates for Flare and Sunburst, and the same as the 2.5 models: image tokens at $8.00 per million in and $30.00 per million out, cached image input $2.00, text input $5.00 and cached text input $1.25. There is no per-image price on that page, so the cost of a GIF or a high-quality image depends on the tokens it uses. OpenAI points to the cost calculator in its image generation guide for estimates. ChatGPT access comes with the subscription tiers.

Who has shown it working, and what OpenAI says it cannot do

• OpenAI shows two sprite-sheet prompts from OpenArt that ask for a seamless looping chibi office-worker animation in six frames laid out three by two. OpenArt's CEO and co-founder Coco Mao is quoted calling GIFs the native language of social media.
• Figma's Itay Schiff, Creative Director at Figma Weave, says the model is being added to Figma Weave. The startup Town used it through the API to explore animated characters in a work assistant.
• OpenAI's own limits list says complex prompts can take up to 2 minutes, text placement can still be unclear, recurring characters or brand elements can drift between generations, and layout-sensitive compositions can be hard to control.
• You may need to finish API Organization Verification in the developer console before using GPT Image models.

Where it sits in the GPT Image line

The older ChatGPT Images 2.0 page covers the April release. OpenAI's API docs now list GPT Image 2, 2.5 and 2.6 side by side, with 2.5 Sunburst and Flare snapshots dated September 8. The 2.6 release adds GIF, complexity, SVG and markup on top of that line. OpenAI says people now create 4 billion images a week across ChatGPT, Codex and the API, up from the 3 billion it gave in September, according to The Verge. Seailife has not generated a GIF or called the API for this page.