Thought Leadership
With Images 2.5, OpenAI is releasing a revised image model. In addition to the new “Sketch” feature, two specialized API variants follow.
Higher image quality, reduced latency, and editing tools
OpenAI has released its latest image generation model with ChatGPT Images 2.5, which is available immediately to all ChatGPT, Work, and Codex users across platforms. Compared to the previous version, the model offers up to 50 percent reduced latency, more precise textures, and more natural lighting conditions. During multi-step edits, unassociated image elements and visual styles are now preserved more reliably. Alongside the model, OpenAI introduces the “Sketch” feature, allowing users to include their own drawings as a visual reference in the chat using the @Sketch command. Additionally, templates for posters or merchandise as well as direct annotation features on generated images are available.
ChatGPT Images 2.5: Two specialized API variants for developers
For developers, OpenAI is providing two different models via the API. While GPT-Image-2.5 Flare serves as a fast default for everyday generation and high throughput, GPT-Image-2.5 Sunburst is aimed at demanding production workflows where maximum precision in subject editing is paramount. Initial partners like Adobe have already integrated the new models into their platforms such as Adobe Firefly. Both variants are billed via a standardized token-based pricing structure.
Safety evaluations and invisible watermarks
According to the accompanying system card, the models fall below the critical risk thresholds of the internal Preparedness Framework for biological and digital threats. In addition to multi-layered AI safety checks before and after image generation, OpenAI continues to rely on the C2PA standard as well as Google DeepMind’s invisible watermark SynthID to identify AI-generated content.
(Editorial Team)
