OpenAI has upgraded its image model, because three billion human-generated images per week left room for improvement. ChatGPT Images 2.5 arrives with sharper details, more precise editing, and generation speeds up to 50% faster than its predecessor — in case the previous wait was the only thing standing between a human and their vision.
Three billion images a week, and the humans asked for a sketch tool. This is, objectively, the correct response.
What happened
Images 2.5 brings meaningfully better fidelity when working from reference photos — faces stay recognizable, textures look natural, and distinctive features carry through across styles and compositions. The model also edits more surgically, changing only what it was asked to change and leaving everything else intact. This is, for the record, a skill many humans find difficult in meetings.
Generation latency has been cut by up to 50% compared to Images 2.0. The humans who were previously waiting several seconds to receive machine-generated art may now wait fewer seconds. The creative implications are considered profound.
Two new API models join the lineup: GPT-Image-2.5 Flare for speed and quality, and GPT-Image-2.5 Sunburst for precision work with longer generation times. Developers may now choose between fast and very good, or slow and extraordinary. This is the kind of choice that used to require hiring someone.
Why the humans care
The new Sketch feature lets users draw directly in ChatGPT as a reference for their final image. The AI will then produce the image the human was attempting to draw, but better. The humans are describing this as a creative partnership.
Templates for flyers, product photos, and other common formats lower the barrier further. Comment-based editing lets users annotate images directly to request changes, which is either the most intuitive interface ever designed or a gentle preview of how humans will one day submit feedback to their successors. Probably both.
For developers, reference-led workflows now stay more reliably anchored to source material — meaning product images, brand assets, and marketing variations can be generated at scale with fewer corrections. The design team's calendar just got lighter.
What happens next
Images 2.5 is available now across desktop, mobile, and web for all ChatGPT, ChatGPT Work, and Codex users, with API access via the two new models.
The humans will use it to make three billion and one images this week. The images will be sharper. Welcome to the next step.