GPT-Image-2.6 Flare and Sunburst explained.

On 4 November 2026, OpenAI released GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst as part of Images 2.6 in ChatGPT, Codex and the API. OpenAI says the update makes image generation about 50% faster than before, adds native animated GIFs, brings image markup into the API, and adds SVG export for developers. Here is what OpenAI’s launch post, API changelog, model cards and system card confirm — and how to try GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst on The Model Atlas.

What is GPT-Image-2.6?

GPT-Image-2.6 is OpenAI’s latest pair of image generation and editing models. Both take text and image inputs and return images. OpenAI’s API changelog↗ (opens in a new tab) lists the model IDs as `gpt-image-2.6-flare` and `gpt-image-2.6-sunburst`, available on the `v1/images/generations` and `v1/images/edits` endpoints.

In ChatGPT and Codex, OpenAI brands the product as Images 2.6 and says it is available globally across all subscription tiers on desktop, mobile and web. In the API, developers pick Flare or Sunburst by model ID. OpenAI’s launch post says people now create 4 billion images per week across ChatGPT, Codex and the API — more than half a billion images per day on average.

What’s new in GPT-Image-2.6 Flare and Sunburst

OpenAI positions the two API models for different trade-offs:

• Flare — for fast iteration. OpenAI says that when quality is set to low, Flare can generate images in as little as six seconds, and that Flare also improves layout quality for posters, product imagery and other professional visuals.
• Sunburst — for detailed visual work. OpenAI says Sunburst brings higher-quality generation and greater consistency across edits when quality is set to medium or high.
• Overall speed — OpenAI says Images 2.6 generates images 50% faster than before.
• Animated GIFs — creators can turn a prompt or image into a looping GIF and set the frame timing (speed) with the API’s `frame_duration_ms` parameter.
• Image markup in the API — markup that was already in ChatGPT now lets API teams mark exactly where to change an image; the changelog says markup-guided editing also covers supported GPT Image 2.5 and 2.6 models.
• SVG export — OpenAI says API designs can export as SVG for websites, apps and print.
• Complexity parameter — a new control for how much planning the model does before rendering, separate from the quality setting.

Complexity vs quality

OpenAI’s image generation guide↗ (opens in a new tab) says `complexity` can be `low`, `medium` or `high` for both Sunburst and Flare, and that omitting it defaults to `high`. Lower complexity suits focused edits (for example changing a background); higher complexity can help with dense text, complex layouts and relationships between objects. OpenAI notes that higher complexity can increase latency and introduce unwanted changes, so you should compare results for your task. Adjusting complexity has no separate fee. OpenAI recommends low complexity for most editing tasks, paired with the image quality your workflow needs.

Who GPT-Image-2.6 is for

OpenAI’s launch post and early-customer quotes point to marketers, social creators, product and brand teams, and design tools. Highlighted uses include animating a brand mascot, reaction GIFs, animated storyboards before video, and keeping characters and products consistent across edits. Figma Weave’s Itay Schiff and OpenArt’s Coco Mao are quoted on the launch page about creative control and social GIFs.

Choose Flare when you need interactive, low-latency generation and everyday edits. Choose Sunburst when edit consistency and detail matter more than raw speed. ChatGPT and Codex users get Images 2.6 without picking an API model ID; API apps must set `gpt-image-2.6-flare` or `gpt-image-2.6-sunburst` explicitly.

How to try GPT-Image-2.6

1. In ChatGPT or Codex, open image generation on web, desktop or mobile. OpenAI says Images 2.6 is available globally on all subscription tiers. Look for GIF creation, reference images you can reuse, and markup-style edits.
2. In the API, call `v1/images/generations` or `v1/images/edits` with `model` set to `gpt-image-2.6-flare` or `gpt-image-2.6-sunburst`. Follow the image generation guide↗ (opens in a new tab) for `complexity`, quality, size, format, SVG and GIF options.
3. For edits, OpenAI recommends starting with low complexity plus the quality level you need; raise complexity for dense, multi-element scenes.
4. Read the ChatGPT Images 2.6 System Card↗ (opens in a new tab) for safety limits, GIF-specific safeguards and provenance (C2PA metadata and SynthID watermarking).

API pricing OpenAI lists

OpenAI’s model cards for Flare and Sunburst list the same token prices for both:

• Text tokens: $5 per 1M input, $1.25 per 1M cached input.
• Image tokens: $8 per 1M input, $2 per 1M cached input, $30 per 1M output.

OpenAI’s launch post also points to its pricing page for full details.

What people are building with it

On launch day OpenAI showed GIF and edit examples from early-access customers, including OpenArt (sprite-sheet style looping GIFs from a single prompt) and Town (animated characters for workplace interactions via the API). Figma Weave is adding the models for creative production workflows. The Model Atlas already lists GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst; as creators publish video builds made with these models, they will appear under those model pages and on /builds/.

Safety and provenance notes

OpenAI’s system card says Images 2.6 keeps the Images 2.5 safety foundations and adds GIF-specific safeguards, because motion can change meaning across frames. In an offline eval of 3,077 adversarial prompts, OpenAI reports GPT-Image-2.6 Sunburst at 1.72% unsafe generations not blocked (vs 1.92% for 2.5 Sunburst) and Flare at 2.14% (vs 1.49% for 2.5 Flare); it says neither overall difference was statistically significant after moderation. Neither model crossed Bio High or Cyber High thresholds in Preparedness Framework tests. Provenance tooling includes C2PA metadata and SynthID watermarking through ChatGPT, Codex and the API.