ChatGPT Images 2.5 Ships With Up to 50% Lower Generation Latency, Plus Sketch and Templates

On September 8, 2026 OpenAI introduced ChatGPT Images 2.5, described as its new state-of-the-art image model bringing sharper details, more precise editing and faster generation. The scale context in the post: people create more than 3 billion images every week across ChatGPT Images and the GPT-Image models in the API.

Model improvements OpenAI lists#

• Image generation latency cut by up to 50% compared with Images 2.0.
• More natural lighting and richer textures, with better preservation of subjects from reference photos.
• Multi-turn editing consistency: earlier changes are more likely to survive, so each edit builds on previous work without rebuilding the whole asset or visibly degrading quality over time.
• Edits change only what was asked, and complex layouts including transparent backgrounds are handled better, with more accurate real-world information in the result.

Three new features in ChatGPT#

• Sketch: draw directly inside ChatGPT and use the rough art as a visual guide, invoked by typing "@Sketch".
• Templates: starting points for popular formats such as posters and merch instead of a blank canvas.
• Comments and prompt sharing: place comments directly on an image for focused edits, and optionally include the prompt when sharing an image so others can rerun the idea with their own details.

API and availability#

Images 2.5 began rolling out on September 8 to ChatGPT, ChatGPT Work and Codex users across all tiers, on desktop, mobile and web. For developers, two models arrive in the API: GPT-Image-2.5 Flare, positioned as the default choice for most applications and said to deliver higher-quality images than GPT-Image-2 at 50% lower latency, and GPT-Image-2.5 Sunburst, built for premium visual workflows needing tighter control across edits, with the post noting it trades longer generation times for extra precision.

Customer quotes carry additional numbers — Manus says Flare delivers high-quality images at two to four times the speed of GPT-Image-2, while Adobe and Runway highlight speed and resolution consistency respectively. These are vendor-stated evaluations; OpenAI's post does not state per-image prices in the body text.

Provenance and safety#

OpenAI says Images 2.5 keeps prompt- and image-level checks to help prevent harmful outputs, and continues to use C2PA metadata and invisible watermarking so images made with its tools can be identified, with further evaluation detail in the system card. The practical shift in this release is editability rather than resolution: latency, multi-turn consistency and the reliability of localized edits are the only dimensions OpenAI quantifies.