OpenAI has released ChatGPT Images 2.5, a new version of the image generation model built into ChatGPT, saying it produces sharper, more realistic images and edits them faster than the previous version.
The company says the update delivers better subject preservation when working from a reference photo, more natural lighting and textures, and roughly 50 per cent lower generation latency than Images 2.0. It has also improved the model’s ability to make precise edits to one part of an image while keeping the rest unchanged across several rounds of back-and-forth editing.
New tools ship alongside the model: a sketch tool lets a user draw a rough visual reference directly inside ChatGPT for the model to work from, pre-built templates cover formats like posters, product photos and merchandise, and a comments feature lets people mark up specific spots on an image to direct an edit. Users can also share a prompt so someone else can recreate the same concept using their own photos.
On the developer side, OpenAI released two API models: GPT-Image-2.5 Flare, the default option, which the company says matches the quality gains at half the latency of GPT-Image-2, and GPT-Image-2.5 Sunburst, a premium option aimed at production creative work that needs tighter editing control. API pricing is posted on OpenAI’s developer platform.
The update is rolling out now to ChatGPT, ChatGPT Work and Codex users across desktop, mobile and web, on all tiers. OpenAI says every image carries C2PA metadata and an invisible watermark, and that prompts and images are checked against its safety systems before generation.
OpenAI’s announcement includes comments from partners Adobe, Runway, Manus and Higgsfield AI, who it says praised the model’s editing precision, speed and image quality.
Source: “Introducing ChatGPT Images 2.5”, OpenAI, 8 September 2026.
