On September 8, 2026, OpenAI announced a new image generation model, "ChatGPT Images 2.5." The company stated that the model reduces image generation time by up to 50% compared to "Images 2.0," which introduced 2K resolution and two levels of inference.
In terms of functionality, consistency in editing through multi-turn interactions has been improved, making it possible to correct details while maintaining the characteristics of the subject. Additionally, a "Sketch" feature, which allows hand-drawn sketches created directly within ChatGPT to be used as references for generated images, and template functions designed for specific formats have been added.
For the developer API, two models will be offered: "Flare," which emphasizes speed and load resilience, and "Sunburst," which specializes in precise control, such as changing only the background. As of now, the API pricing has not been disclosed.
ChatGPT Images 2.5 will be rolled out sequentially across all ChatGPT plans, including the free plan, as well as in Codex. Generated images will include C2PA metadata indicating they were AI-generated, as well as "SynthID," an electronic watermark resilient to modifications such as resizing.
Source:
- OpenAI が ChatGPT Images 2.5 を発表、画像生成が50%高速に - Pasquale Pillitteri (Google News: OpenAI, 2026-09-09)