SAN FRANCISCO, September 16 — OpenAI has unveiled ChatGPT Images 2.5, a new image generation model the company says sharpens image quality, editing, speed and creative control, in an update that reaches ChatGPT, ChatGPT Work and Codex users across desktop, mobile and web. The company announced the model on September 8, 2026.
Users already create more than three billion images each week across ChatGPT Images and the GPT-Image models in OpenAI’s API, according to the company — a scale that explains why even incremental gains in speed and control matter to the people who rely on these tools daily.
Sharper detail, faster turnaround
OpenAI says the new model delivers sharper details and richer textures, produces more natural lighting, preserves reference photos more accurately during image creation, and follows editing instructions more reliably across multiple editing turns. Generation latency has dropped by up to 50% compared with Images 2.0. The central change involves editing.
ChatGPT Images 2.5 can make focused changes without altering unrelated details, letting users modify specific parts of an image while keeping the composition intact — something OpenAI says can help with products, backgrounds, text and other visual elements. The model also improves consistency during longer editing sessions, so earlier changes are more likely to remain intact and each new edit can build on the previous version without reducing image quality.
For anyone who has spent an afternoon coaxing a single image into shape, that consistency is the headline. OpenAI says the improvements could benefit users working on detailed creative projects as well as marketing teams and businesses creating repeated visual assets.
A sketch pad inside ChatGPT
OpenAI is also introducing a feature called Sketch that lets users draw directly inside ChatGPT. Users can create a rough drawing and use it as a visual reference, and ChatGPT can then turn that sketch into a more complete image. The feature is intended to help users communicate ideas without creating detailed artwork first, and OpenAI says users can sketch layouts, clothing concepts, room designs or simple visual ideas.
Users can access the feature by typing “@Sketch” in ChatGPT. The update adds templates for popular creative formats, with options such as posters and merchandise designs, reducing the need to begin with a blank canvas.
Users can then add information, design elements and preferred visual styles. OpenAI says ChatGPT Images 2.5 also improves its understanding of complex visual instructions and can handle detailed layouts and transparent backgrounds more effectively.
New models for developers
Through its API, OpenAI has introduced two new models: GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst. Flare focuses on quality, editing and speed, and OpenAI says it delivers higher-quality images than GPT-Image-2 with 50% lower latency. Sunburst targets premium creative workflows and provides tighter control for detailed editing and production-ready visual work.
OpenAI says these models can support creative, marketing, retail, media and other business applications. On safety, OpenAI says it is continuing its image safety measures, including prompt and image checks designed to reduce harmful outputs, and that it uses C2PA metadata and invisible watermarking to help identify images created with its tools.
What to watch next is how quickly the editing-consistency claims hold up once marketing teams and independent creators put the model through long, multi-turn sessions — and whether the two API models shift how businesses build repeated visual assets at scale.
