ChatGPT Images 2.5 Cuts Generation Latency by Half

What Changed in OpenAI's New Image Model

OpenAI released ChatGPT Images 2.5 on September 8, calling it the company's new state-of-the-art image model. The headline improvement is not a new capability but a speed one: OpenAI says image generation latency is down by as much as 50% against Images 2.0. Alongside the ChatGPT release, two companion models arrived in the API, gpt-image-2.5-flare and gpt-image-2.5-sunburst.

• Up to 50% lower latency compared with ChatGPT Images 2.0
• Available to every tier — ChatGPT, ChatGPT Work and Codex users on desktop, mobile and web
• Two new API models: gpt-image-2.5-flare and gpt-image-2.5-sunburst
• Sketch input arrives in the ChatGPT interface, invoked with an @Sketch mention, alongside templates and prompt sharing

The quality claims centre on control rather than raw fidelity. OpenAI describes more natural lighting and richer textures, better preservation of subjects from a user's reference photos, and more reliable multi-turn editing — the model changes only what it was asked to change and leaves the rest of the image alone, even with complicated subjects and backgrounds.

Why Editing Precision Matters More Than Fidelity

Anyone who has tried to iterate on a generated image knows the actual failure mode. You ask for one small change, and the model quietly redraws the face, shifts the lighting, and moves a background object three pixels left. Each round of edits drifts further from the thing you liked in the first place. That is a workflow problem, not an aesthetics problem, and it is the one Images 2.5 is aimed at.

OpenAI also cites better handling of complex layouts, including transparent backgrounds, and more consistent adherence to a requested visual style across a series of images. Style consistency is the quiet prerequisite for using generated art in anything that has to look like a set — a product line, a comic, a slide deck. The previous Images 2.0 release in April focused on thinking mode and text rendering; this one is about keeping a good result once you have it.

What Is the Sketch Tool in ChatGPT Images 2.5?

Sketch is the interface change most likely to alter how people actually work. Rather than describing spatial arrangement in a paragraph of prose — the subject slightly left of centre, the horizon two-thirds up — a user can draw a rough layout and hand it to the model as structural input, invoked with @Sketch. Pre-formatted templates and prompt-sharing tools round out the update, the latter letting someone re-run another person's generation logic against their own source assets.

Where This Sits in the 2026 Image Model Race

Image generation has become the least glamorous and most-used part of the frontier model stack. The competitive pressure this year has been on latency and editability rather than benchmark scores, because those are the properties that decide whether a model gets used in production. A halving of generation time changes what is practical inside an interactive loop, and it lowers the cost of the iterate-until-right pattern that most real work depends on.

It also lands in a busy stretch for OpenAI, arriving days after the GPT-6 Astra rollout to ChatGPT Plus. For teams building on the API, the practical decision is now between two named variants rather than a single endpoint, which suggests a speed-versus-quality split of the kind that has become standard across model families. More in our AI coverage.

Sources: OpenAI — September 8, 2026; Axios — September 8, 2026; 9to5Mac — September 8, 2026.