OpenAI ships ChatGPT Images 2.5, cutting generation latency up to 50% and adding a sketch-to-image tool
The update ships two new API models, Flare and Sunburst, and rolls out to every ChatGPT tier, including free accounts.
OpenAI released ChatGPT Images 2.5 on September 8, an update to its in-app image generator that the company says "reduced image generation latency by up to 50% compared with Images 2.0." OpenAI also claims sharper detail, more natural lighting and texture, better preservation of subjects pulled from reference photos, and more reliable instruction-following across multiple rounds of edits in the same conversation.
What's new beyond speed
The headline addition is Sketch, a tool that turns a drawing made inside ChatGPT into a reference image the model builds from. That workflow didn't exist in 2.0. OpenAI paired it with new templates for common image formats, inline comments on generated images, and prompt sharing, which lets others reuse one user's prompt. For developers, the release ships two API models: GPT-Image-2.5 Flare, the default, optimized for the lower latency figure, and GPT-Image-2.5 Sunburst, which trades generation time for more precise editing control.
Free accounts get it too
Images 2.5 is available now across ChatGPT, ChatGPT for work, and inside Codex, on web, desktop and mobile, and on every ChatGPT plan. That differs from OpenAI's other major release this month: GPT-6 Astra, which launched September 3 and whose demand OpenAI says has since forced it to pause new sign-ups to the $200 Pro tier entirely. Shipping a flagship image upgrade to free users the same week compute is tight enough to close a paid tier suggests OpenAI treats image generation and Astra's heavier compute features as separate capacity pools.