OpenAI has released an updated version of its image generation technology under the name ChatGPT Images 2.5. According to the company, the ChatGPT Images and GPT‑Image models in the API have been used to generate more than 3 billion images in total. The new release is said to improve multi-turn instruction following, deliver faster responses, and better preserve the subjects in reference photos.
New models in the API
Two new model IDs are now available in the OpenAI API:
- gpt-image-2.5-sunburst
- gpt-image-2.5-flare
The author of the report indicates a preference for Sunburst as the stronger option for precision editing. Their recommendation is to choose Sunburst when editing accuracy matters most, and Flare when you need fast, high-quality everyday image generation.
Developer tooling and example usage
The author updated their openai_image.py command-line tool to accept one or more reference images. The example command they shared is:
uv run https://tools.simonwillison.net/python/openai_image.py
'add a raccoon scientist studying the chart thoughtfully'
-i https://static.simonwillison.net/static/2026/openai-agent-usage.webp
-m gpt-image-2.5-sunburst
This demonstrates using a reference image and the Sunburst model to add a new element to an original photo — in this case, a raccoon scientist thoughtfully studying a chart.
Why this matters
Improvements in instruction following and in-preservation of reference subjects are relevant for users and developers who require precise image edits, such as product photography, illustrations, or other visual content workflows. Offering two models that prioritize either precision (Sunburst) or speed (Flare) lets creators and engineers pick the model that best fits their workflow.
The usage figure provided by OpenAI (more than 3 billion images) indicates broad adoption, and the 2.5 release appears to be a targeted upgrade for that existing user base.
