On September 8, 2026, OpenAI officially released its most advanced AI image-generation model to date, ChatGPT Images 2.5. Compared with Images 2.0 launched in April 2026, the new version significantly improves image detail sharpness, natural lighting and texture richness, while reducing image generation latency by up to 50%, making creation and iteration more efficient and fluid. The model has been rolled out to all ChatGPT, ChatGPT Work and Codex users on desktop, mobile and web, and two API models are available to developers.

Details: comprehensive upgrades in quality, speed and editing capability

OpenAI says more than 3 billion images are now generated weekly via ChatGPT Images and the APIs. The core improvements in Images 2.5 focus on visual quality and interactive experience:

Visual improvements

Images produced by the new model exhibit sharper details, more natural lighting and richer textures. In reference-photo–driven generation and editing, the model preserves subject characteristics such as faces and pets much better, reducing previously common issues like face distortion or detail drift. In multi‑round editing, the model follows instructions more reliably, modifying only specified areas while maintaining consistency of the subject, composition and background.

Significant speed optimizations

Compared with Images 2.0, generation latency is reduced by up to 50%. This allows users to produce initial drafts and iterate much faster, bringing the creation workflow closer to real-time feedback. The improvement is particularly important for design, marketing and content-creation scenarios that require extensive trial and error.

New tools and workflow enhancements

Image comment editing: Users can add comments directly on images to precisely indicate areas to change, enabling more focused local edits.
Sketch function: Users can draw sketches directly within ChatGPT as references, lowering the barrier of pure text prompts.
Prompt sharing: Generated images can include the original prompt, making it easier for others to recreate ideas using their own photos or parameters.
Template support has been expanded to cover common formats such as posters and merchandise.

Developer APIs: two models

GPT-Image-2.5 Flare: The default recommended model, providing quality improvements while delivering roughly 50% lower latency, suitable for most use cases including social media content, product prototypes, visual search and bulk generation. Improved transparent-background support helps with brand assets and UI element creation.
GPT-Image-2.5 Sunburst: Targeted at creative and professional workflows requiring higher precision; it takes longer to generate but offers superior control over details and complex edits, suitable for commercial advertising and finely detailed product imagery.

On the safety side, OpenAI continues to apply C2PA metadata and invisible watermarking to help identify AI-generated content and reduce risks of unauthorized reuse and deepfakes.

Technical background and product positioning

Images 2.0 introduced capabilities such as 2K resolution, multiple aspect ratios, web realtime information retrieval and non-Latin language support. Images 2.5 builds on that foundation with a focus on "faster, more accurate, more usable," especially strengthening reference consistency and multi‑round edit stability. This aligns with the trend in generative AI shifting from "can generate" to "can precisely control and efficiently iterate."

The model serves both everyday creators and developer product integrations, reflecting OpenAI's strategy to deeply embed image capabilities within the ChatGPT ecosystem and API