OpenAI launched Images 2.6 on November 4, 2026 in ChatGPT, Codex and the API. For a small or mid-size team that already leans on ChatGPT for product shots, social assets and mockups, the practical change is speed, motion and tighter edit control. It is not a new subscription product.
OpenAI says Images 2.6 generates images 50% faster than before. In ChatGPT and Codex it is live globally across all subscription tiers on desktop, mobile and web. Designers do not need a new plan to try it. They do need a workflow choice: stay in ChatGPT for one-off visuals, or wire the API into your own tools.
Flare for drafts, Sunburst for the keepers
In the API, OpenAI ships two models with the same token prices and different jobs. GPT-Image-2.6 Flare is positioned for fast iteration. OpenAI says that with quality set to low, Flare can return an image in as little as six seconds. It also says Flare improves layout quality for posters and product imagery.
GPT-Image-2.6 Sunburst is for detailed visual work. OpenAI positions it for stronger consistency across edits when quality is medium or high. A workable split is to draft in Flare, then switch the model string to Sunburst for finals and brand-sensitive edits.
Both models cost $5 per 1M text input tokens ($1.25 cached) and $8 per 1M image input tokens ($2 cached). Image output is $30 per 1M tokens, per those model pages. Neither lists Batch support. Both are limited to the Image API generation and edit endpoints.
OpenAI has not published a fixed per-image price for 2.6, so cost still depends on output tokens per render.
GIFs, SVG, markup and complexity
The ops-facing features sit in the image generation guide. Animated GIFs and SVG exports come through the Image API generation endpoint. Markup-guided edits, already in ChatGPT, now land in the API. A designer can mark where a change should happen instead of rewriting the whole prompt.
A new complexity parameter accepts low, medium or high, and defaults to high. It controls how much planning the model does before rendering. OpenAI says changing complexity has no separate fee. Quality still controls visual fidelity.
GIF output accepts 2 to 64 frames (default 9) and 20 to 2,000 milliseconds per frame. SVG suits icons and editable graphics. The edit endpoint does not support SVG or GIF output.
If your agency ships AI-made assets into the EU, older generators must mark machine-made output by 2 December. We covered that duty in our EU AI Act content marking deadline piece. Motion and SVG do not remove it.
OpenAI also claims people now create 4 billion images per week across ChatGPT, Codex and the API. Treat that as the company’s own usage figure. We could not independently verify the methodology.
Where it still breaks
Batch is unsupported on both 2.6 models. High-volume overnight catalogues still belong on GPT-Image-2.5 if you need the Batch discount. Responses and Chat Completions do not support 2.6 either. You call v1/images/generations and v1/images/edits.
There is no dated snapshot to pin yet, only the bare model IDs. If you are building a looping tool rather than a chat, revisit how an AI agent differs from a chatbot before you bolt image calls onto an automated pipeline.
What to do this week
Pick ten recent ChatGPT image jobs from your team. Re-run five drafts on Flare at low quality and five keepers on Sunburst at medium or high. Log latency and output tokens from the API responses. Then decide whether 2.6 belongs in ChatGPT only, in your production Image API path, or both.
Keep 2.5 Batch for any bulk job until OpenAI adds Batch for 2.6.
Sources
Get the weekly email
The week's most useful AI news, a tutorial and an interview.
Anna Jakovleva
Anna Jakovleva is the senior editor for AI tools at AI Observer, leading coverage of agents and automation, design and product, and the site's tutorials. She worked in operations and product teams before joining, so she judges tools by setup time, cost and where they break. Her writing is hands-on and practical.
Spotted a mistake? Tell us on the corrections page.