OpenAI has launched ChatGPT Images 2.5, a new image-generation model that it describes as state-of-the-art. The model is rolling out to ChatGPT, ChatGPT Work, and Codex users on desktop, mobile, and web, and it brings sharper details, faster generation, and more precise editing to users' creative workflows.
Key capabilities include up to a 50% reduction in image-generation latency compared with the previous Images 2.0 model, improved consistency when preserving subjects in reference photos, and more reliable adherence to editing instructions across multiple conversation turns. The model is also better at following complex visual instructions and retaining requested visual direction over time.
Two new API models accompany the release: GPT-Image-2.5 Flare, the default choice for most applications with higher-quality images than GPT-Image-2 at 50% lower latency, and GPT-Image-2.5 Sunburst, aimed at premium creative workflows that require tighter control across edits. OpenAI says the same fidelity improvements make reference-led workflows more reliable for developers building image generation into creative, marketing, retail, and media products.
New product features in ChatGPT include Sketch, which lets users draw directly in ChatGPT to use as a reference for their final image, and Templates for popular formats like flyers and product photos. Users can also place comments directly on images for focused editing and share the prompts behind their images so others can iterate on similar ideas.
On safety, OpenAI says Images 2.5 builds on existing safeguards with added protections for new risks that arise as models become more capable. The safety stack includes prompt-layer and image-layer checks, a safety-focused multimodal reasoning model that monitors inputs and outputs, and continued use of C2PA metadata and invisible watermarking to help identify generated images. According to an OpenAI safety evaluation using adversarial prompts, GPT-Image-2.5-Sunburst blocked 98.91% of unsafe generations and GPT-Image-2.5-Flare blocked 98.59%, both improving on the 98.36% block rate of ChatGPT Images 2.0.
Early customer feedback highlighted speed and editing accuracy: Adobe noted the models support faster generation and resolution consistency, while Runway and Manus cited lower latency and quality for creative and brand-asset workflows.