ChatGPT Images 2.5 Launches With Faster Generation, Smarter Editing and Two New API Models

On the morning of 8 September 2026, anyone who opened ChatGPT on their phone or laptop found something quietly different. Ask it to generate an image and the result arrived noticeably faster. Ask it to tweak just the background, or the lighting, or a face — and it held together in a way earlier versions often didn’t. OpenAI had pushed out ChatGPT Images 2.5, and the company wasn’t shy about the billing: “faster, sharper, smarter.”

The headline number is a reduction in image generation latency of up to 50% compared with the previous Images 2.0 model. That’s the difference, in practice, between waiting and not waiting — the kind of gap that changes whether a tool feels useful or frustrating. OpenAI framed the release as its best image generation model yet, and the rollout covered all ChatGPT tiers from the outset: free users, paid subscribers, ChatGPT Work accounts, and Codex users, across desktop, mobile and web, with no separate waitlist.

What’s Actually New in Images 2.5

Speed aside, the more telling changes are in editing. The new comment-based editing tool lets users click on a specific region of an image and leave a note — “make this darker,” “remove this object,” “change the jacket colour” — rather than rewriting an entire prompt from scratch. It’s the difference between scrawling notes in the margin of a document and having to retype the whole thing.

The Sketch tool goes further, letting users draw directly inside ChatGPT to indicate what they want changed or added. And a set of pre-built templates — covering posters, merchandise and other common formats — means someone running a small business or community event can get a usable visual without starting from a blank prompt.

OpenAI also reported improvements in natural lighting, texture rendering, transparent backgrounds, and what it describes as better preservation of faces, objects and overall composition across repeated edits. Multi-turn consistency — keeping an image coherent across several rounds of changes — has been a persistent weak point in AI image tools, and the company says Images 2.5 addresses this directly. These claims are vendor-reported, and independent benchmarking by external researchers will take time to catch up.

Flare and Sunburst: Two Speeds for Developers

For developers building on the API, OpenAI has split the 2.5 generation into two distinct models. GPT-Image-2.5 Flare is built for speed, delivering quality comparable to the previous GPT Image 2 model but faster. GPT-Image-2.5 Sunburst trades some of that speed for higher precision and finer detail — suited to workflows where quality matters more than turnaround time.

It’s a sensible split. A social media scheduling tool needs something different from a professional design platform, and offering both within the same generation gives developers more control over that trade-off. Pricing for both API models is reportedly aligned with previous GPT-Image-2 rates, denominated in US dollars; exact GBP figures will vary with exchange rates and regional pricing policies, and no official UK-specific pricing has been confirmed.

The maximum supported resolution remains up to 4K, and the model continues to accept up to 16 reference images — both specifications carried over from GPT Image 2 rather than expanded in this release.

The Broader Debate Around AI Image Tools

Not everyone greets the update with straightforward enthusiasm. Designers and illustrators have watched AI image tools move from novelty to serious competition in a relatively short time, and a faster, more controllable model accelerates that pressure.

There are also unresolved questions about copyright and provenance. When an AI generates a commercial image, who owns it? How was the training data assembled? OpenAI has not published specific environmental metrics for Images 2.5, and critics note that more capable models generally demand more compute — which carries an energy cost that hasn’t been publicly quantified for this release.

Sam Altman, OpenAI’s chief executive, has previously described the company’s image tools as part of a broader push to make creative production accessible to people without specialist skills. But that accessibility cuts both ways. Faster, more realistic synthetic images also lower the barrier to misleading visuals, and regulators in the UK and elsewhere are still working out what labelling and provenance standards should look like.

Availability and What to Expect

The rollout is already under way. Users on iOS, Android and web should see Images 2.5 as the default image generation option within ChatGPT. The Sketch tool and comment-based editing appear in the same interface, so there’s nothing additional to install or unlock. Templates and prompt sharing are also available from launch, making this a notably broad update rather than a feature drip.

But the real test will come over the coming weeks, as independent researchers and everyday users put the model through its paces in ways that go beyond what a launch announcement can capture.

What This Means for Kent Residents

Anyone in Kent using ChatGPT — on the free tier or a paid plan — gets access to Images 2.5 from the same rollout as users anywhere else in the world, so faster generation and the new editing tools are available now across the county. For local small businesses, independent traders, and community groups looking to produce posters, social media graphics or event materials without hiring a designer, the lower turnaround time and template tools could reduce both cost and effort. Schools, colleges and creative businesses across Kent using AI image tools in professional or public-facing work should remain mindful of UK GDPR obligations, emerging AI governance guidance from UK regulators, and existing intellectual property rules around commercially used AI-generated imagery.