Back

OpenAI's ChatGPT Images 2.5 leads on edits but trails on realistic first images in side-by-side tests

Official ChatGPT Images 2.5 promotional still with sculpted letterforms

OpenAI says ChatGPT Images 2.5, launched Sept. 8, 2026, makes sharper edits than Images 2.0 and cuts latency, the delay before it responds, by up to about 50%. In side-by-side tests by creators, a rival model, Nano Banana Pro, made better starting photos, while Images 2.5 did better at edits that combine several reference images. A common approach among the testers is to make the first image elsewhere and refine it with Images 2.5.

OpenAI launched ChatGPT Images 2.5 on Sept. 8, 2026. The company says it brings sharper detail, more precise editing and up to about 50% lower latency than Images 2.0, along with a Sketch feature, templates and two versions for developers, Flare and Sunburst, offered through an API, a connection that lets developers plug a service into their own apps. For creators who already choose tools by how well they edit versus how well they make a first image, those claims point to a practical split.

In side-by-side tests by independent creators, the Sunburst version, run through Magnific, lost to Nano Banana Pro on starting photos. It beat Nano Banana Pro on combining several reference images into one picture, outfit edits, swapping billboards and text, matching styles and replacing products. The comparison comes from creators' own testing, not from OpenAI.

Testers reported that faces seen up close can show fractal-like or oversharpened texture, that images became oversharpened over repeated rounds of edits, and that posters still needed their lettering cleaned up.

A common approach is to make the starting image elsewhere, then use Image 2.5 for controlled changes. The testers put the cost of a 4K image, a very high-resolution picture, at about $1 on Magnific.

  • @curiousrefuge Video

    A filmmaking review ran Sunburst via Magnific against Nano Banana Pro and others. Starting photos favored Nano Banana Pro. Multi-reference edits favored Image 2.5.

    Watch on YouTube

More on OpenAI