OpenAI announced ChatGPT Images 2.5 on September 8, 2026, describing improvements in detail fidelity, editing precision and generation speed. The company said weekly image creation across ChatGPT Images and GPT‑Image API models exceeds 3 billion, and that Images 2.5 reduces generation latency by up to 50% compared with Images 2.0.
Key updates
According to OpenAI, Images 2.5 delivers sharper details, more natural lighting and richer textures, and preserves subjects from reference photos more reliably. The model is said to follow editing instructions more consistently across multiple turns, reducing degradation when a sequence of edits is applied.
OpenAI described two API variants: GPT‑Image‑2.5 Flare, positioned as the default for most applications and said to produce higher-quality images than GPT‑Image‑2 at about 50% lower latency; and GPT‑Image‑2.5 Sunburst, which OpenAI positions for premium workflows that require tighter control and allow for longer generation times.
New tools and workflow features
The release also adds product features in ChatGPT to support creative workflows. Sketch is a new tool that lets users draw inside ChatGPT as a visual guide; OpenAI noted users can invoke it by typing “@Sketch”. Templates for common formats such as posters and product photos are offered to accelerate starts, and users can place comments directly on images and share the prompts used to create them so others can reproduce or adapt ideas.
OpenAI said Images 2.5 improves targeted editing—changing a single element while preserving composition and brand treatment—and handles more complex layouts including transparent backgrounds. The company stated the model better retains requested visual direction over a series of edits, which it said matters for on‑brand creative assets and UI concepts.
Safety, pricing and availability
OpenAI said existing safeguards remain in place, including prompt and image checks, C2PA metadata and invisible watermarking to identify generated content. The company also published an evaluation system card for further details.
Early feedback included a comment from Axultan Alimkulov, Head of Product at Higgsfield AI, who said GPT‑Image‑2.5 Flare understands what not to change and preserves character and composition when making edits. Images 2.5 is rolling out to ChatGPT, ChatGPT Work and Codex users on desktop, mobile and web, and GPT‑Image‑2.5 Flare and GPT‑Image‑2.5 Sunburst are available in the API.
Original source: OpenAI News