OpenAI began rolling out its updated image generation feature, ChatGPT Images 2.5, on September 8, 2026. The release comes about five months after Images 2 debuted in April, and it brings faster generation along with new editing tools. The update is reaching ChatGPT, ChatGPT Work, and Codex users across all tiers, on desktop, mobile, and web.

Building on Images 2

Images 2, released in April, added 2K-resolution output, support for multiple aspect ratios, web-search grounding for up-to-date references, stronger support for non-Latin languages, and two processing modes: Instant for speed and Thinking for precision. Images 2.5 builds on that foundation with further gains in quality, speed, and editing accuracy.

A sketch tool and several new features

The most notable addition is Sketch, which lets people draw directly inside ChatGPT and use that drawing as a reference for image generation. A rough sketch is enough to guide the composition of the resulting image.

The update also adds templates for common formats such as posters and merchandise, a commenting tool that lets users pin edit requests to specific spots on an image, and a way to share the prompts behind a generated image with other users. In hands-on testing by one outlet, a single photo was transformed into styles such as film noir, stained glass, and a Renaissance-style portrait while still preserving the subject's distinctive features. A tattoo-design feature lets users pick from and refine multiple design concepts, then connects them with a real tattoo artist to bring the design to life.

Generation is up to 50% faster

On the image-quality side, OpenAI says lighting looks more natural and textures are richer, and the model does a better job preserving the subject of a reference photo. Edits also stay more consistent across multiple rounds of instructions.

As for speed, OpenAI says generation is up to 50% faster than Images 2.0. ChatGPT is now generating more than 3 billion images per week, and the efficiency gains appear to be aimed in part at supporting that scale.

Two new API models for developers

Developers get two new models: GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst. Flare is the default option, delivering higher-quality images than GPT-Image-2 while cutting latency by 50%, and OpenAI positions it for social content, product imagery, visual search, and rapid prototyping. Sunburst is aimed at premium workflows that need tighter control over edits, such as production-ready campaign creative and polished product photography.

API pricing is $8.00 per million tokens for image input ($2.00 when cached), $30.00 per million tokens for image output, and $5.00 per million tokens for text input ($1.25 when cached). On safety, OpenAI reports that 1.09% of Sunburst generations and 1.41% of Flare generations were flagged as unsafe.

Summary

OpenAI rolled out ChatGPT Images 2.5 on September 8, 2026, adding a Sketch drawing tool, comment-based editing, and generation that is up to 50% faster. Developers also gain two purpose-built API models. Coming right after the GPT-6 Astra launch earlier this month, the update underscores how quickly OpenAI has been iterating on its product lineup.

References