OpenAI has introduced ChatGPT Images 2.5, its most advanced image generation model. The new version reduces image generation latency by up to 50% compared to the Images 2 model released in April 2026.
The model introduces a Sketch function, accessible via the '@Sketch' command, allowing users to draw sketches directly in the chat and combine them with style and detail descriptions to create full images.
New templates for common formats like posters and products are now available to lower the entry barrier for creators. Users can also add annotations to images to focus on specific modification areas and include original prompts when sharing.
The in-image comment feature allows users to mark specific locations and issue commands like 'delete this area' or 'change this to blue'. The model modifies only the specified area while keeping the rest of the image unchanged, maintaining consistency and quality during multi-round editing.
OpenAI has released two versions: GPT-Image-2.5 Flare, designed for API applications, creator content, and high-volume generation, and GPT-Image-2.5 Sunburst, intended for high-precision editing workflows, such as marketing materials and refined product images, though it takes longer to generate.
The API supports various resolutions, including 1024x1024, 2048x2048, and 3840x2160 (4K landscape). The maximum length of one side is 3840 pixels, with total pixels ranging from 655,360 to 8,294,400. All outputs include C2PA metadata and invisible watermarks for identification.
