OpenAI shipped ChatGPT Images 2.5 on September 8, 2026. The head-turning stat is speed: image generation is now up to 50% faster than the ChatGPT Images 2.0 model it replaces. That's good news, but the upgrade you'll feel more is the editing. You can now change one detail in a picture without the model quietly redrawing everything else, and edits you've already locked in stop drifting as you keep working.
This is the second major update to ChatGPT's built-in image tool this year. Images 2.0 landed in April with 2K output and a reasoning mode. The 2.5 release doesn't chase higher resolution. It goes after the two things people complained about most: the wait, and pictures that fell apart when you tried to tweak them.
| What Changed | Why It Matters |
|---|---|
| Faster generation | Image generation latency drops by up to 50% |
| Precision editing | Only the region you select changes |
| Reference fidelity | Faces, pets, and products stay recognizable |
| New tools | Sketch, Templates, and image comments |
How Much Faster Is ChatGPT Images 2.5?
OpenAI's official number is "up to 50% lower latency" compared to Images 2.0. Note the words "up to." Not every image will render twice as fast. Simple prompts and smaller canvases gain the most, while complex scenes and multi-subject edits still take longer.
Why does this matter? Because most people don't generate one picture and stop. They generate, look, adjust, generate again. Cut each round in half and that loop gets a lot less painful. It also makes the whole thing feel more like a real editing tool and less like a slot machine you have to wait between pulls on.
If you're on the free ChatGPT tier, the faster model rolls out the same way, but you still run into whatever image-generation limits your plan has.
Precision Editing in ChatGPT Images 2.5
This is the change that fixes the most irritating thing about AI image tools. Before, if you asked the model to turn a white background blue, it might also change the product's color, the lighting, or the composition. One small request, three unexpected edits.
ChatGPT Images 2.5 is built around the idea that you should be able to say "change only this" and have it stick. You can now mark a region on an image, or leave a comment like "remove this" or "make this blue," and the model will touch only that part. Everything else stays where it was.
Developers building on the API get the same benefit. You can swap a product, a background, or a line of text while keeping the subject, composition, and brand treatment around it intact.
Reference Fidelity: Faces and Pets Stay the Same
Another long-standing pain point: you upload a photo of a person or a pet, ask the model to restyle it, and the result looks like a different person or a different cat. The outfit changes, and so does the face.
ChatGPT Images 2.5 handles reference photos better. When you turn a portrait into a retro studio shot or push a pet through film noir and renaissance portrait styles, the subject's distinct features carry through. OpenAI says this holds across lighting, texture, and even character-specific detail.
Early testing from Axios backs this up. Their testers moved a cat photo through several styles and found the model kept the animal recognizable. That's a big deal for anyone making consistent series of product shots or branded visuals, where the whole point is that the recurring element doesn't change on you.
New ChatGPT Tools: Sketch, Templates, and Image Comments
Alongside the model, OpenAI added a few tools to the ChatGPT app that lower the bar for getting started.
- Sketch is the flashiest one. Type "@Sketch" in the chat box and you get a drawing panel. Scribble a rough layout, add a note about style and color, and the ChatGPT image generator turns your doodle into a finished picture. You don't need to be able to draw. You need to be able to describe roughly where things go.
- Templates cover common formats like posters, product shots, and flyers, so you're not starting from a blank prompt every time.
- Image comments let you annotate directly on a picture to point at the exact area you want changed. It's the same idea as leaving a note in Figma, and it's much faster than typing out "the object in the top-left corner."
You can also share a prompt along with an image now, so someone else can reuse the same setup with their own photo.
GPT-Image-2.5 Flare and Sunburst for Developers
If you build on the API, OpenAI split the new model into two versions.
- GPT-Image-2.5 Flare is the default. It's the balanced option, with roughly a 50% latency cut, and it's meant for social content, product visuals, prototypes, and high-volume generation.
- GPT-Image-2.5 Sunburst trades speed for control. It's aimed at final, brand-grade creative work where you'll accept a longer wait for finer detail. Both models are priced the same as the standard 2.0 tier: $8 per million image input tokens and $30 per million output tokens.
The safety layer hasn't changed either. Outputs still carry C2PA metadata and an invisible watermark, so images made with OpenAI's tools stay traceable.
- Bagaimana untuk memuat turun Gacha Nox pada Android & iOS?
- Busuu: Yabancı dil öğren ücretsiz olarak nasıl indirilir?
- How to Download Tile Family®:Match Puzzle Game APK Latest Version 2.09.0 for Android 2026
The 2.5 update is free across the ChatGPT app, the web client, and ChatGPT Work, and it doesn't raise your existing image-generation limits. If you mostly drop in casual prompts and take whatever comes out, the speed bump is a nice bonus. If you edit pictures, run a shop, or make social content, the precision editing and reference fidelity are the real reason to update.
ChatGPT Images 2.5 is rolling out now on Android, iOS, and the web. To get the latest version, just download ChatGPT APK from APKPure. It typically updates within hours of the official release.