Loading page…
The update turns image generation toward a revision workflow, but its central promises—speed, fidelity and consistency—are OpenAI’s own performance claims.
Listen to this story
OpenAI is rolling out ChatGPT Images 2.5 across ChatGPT, Work, and Codex, with a focus on localized edits that preserve the rest of an image. It claims up to 50% lower latency than Images 2.0, but results may vary by request and remain company-reported. New Sketch, templates, and reusable prompt sharing broaden the workflow. API users get two variants: faster, general-purpose Flare and slower, more controllable Sunburst. The key question is whether preservation holds across long, complex editing sessions.
Images 2.5 supports direct comments on image regions, aiming to change one element without disrupting composition or background.
Flare is OpenAI’s default API recommendation; Sunburst targets premium workflows requiring tighter edit control and accepts longer generation times.
Templates support posters, flyers, merchandise, and product photos, but were not yet available in ChatGPT Work.
OpenAI has released ChatGPT Images 2.5, an image-generation upgrade built around a difficult promise: make a local change without forcing users to rebuild the rest of an image. The company says the release improves editing consistency and cuts generation latency by up to 50% versus Images 2.0, though those performance results are OpenAI’s own claims.
OpenAI says Images 2.5 better preserves subjects from reference photos, follows complex visual instructions, retains styles and maintains earlier changes through multiple rounds of editing. Those are useful targets for iterative creative work, but they remain company-reported performance claims.
The speed claim is equally central. OpenAI says Images 2.5 has up to 50% lower image-generation latency than Images 2.0, reducing the wait between an unsuccessful result and the next revision.
Users can place comments directly on an image to identify a region and describe a requested change. OpenAI says the model is better at changing that element while preserving the surrounding subject, composition and background—even in more complex scenes.
Images 2.5 is rolling out across ChatGPT, ChatGPT Work and Codex on desktop, mobile and web. Access is not identical across those products: OpenAI’s release notes say templates were not yet available in Work mode.
OpenAI also released GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst through its API. Flare is the company’s default recommendation for most applications and is described as higher quality than GPT-Image-2 at 50% lower latency. Sunburst is for premium visual workflows that need tighter control across edits and take longer to generate.
OpenAI says the release continues prompt and image checks, C2PA metadata and invisible watermarking to help identify images made with its tools. Its system card says the GPT-Image-2.5 series generally performed on par with or better than Images 2.0 in the company’s safety evaluations, while minor regressions were not statistically significant.
The release gives users more ways to guide an image before and after generation. Whether its promised preservation holds through long editing sessions is the quality question the new workflow puts at the center.
Story updates
omg, we have image AGI. forget arc-agi, chatgpt image 2.5 just passed the letter 'r' benchmark. image 2.0 couldn't circle the letter 'r' but 2.5 can.
Editorial analysis
This release packages model improvements with interface changes that make image creation look more like a sequence of revisions than a single prompt. The consequential test is whether the model can preserve the parts of an image users did not ask to change, especially over multiple edits. OpenAI’s launch claims that it can, but the company has not supplied an independent measure of that behavior here. The API split is also revealing: Flare prioritizes lower latency, while Sunburst trades longer waits for tighter control. Watch whether teams adopt that choice as a real workflow division or find one option sufficient for most work.
Loading discussion...
Reader poll
Tell us which priority matters more in your work.
Results are available to signed-in participants. Vote or post a comment to unlock them.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.