ChatGPT Images 2.5: Faster generation with Sunburst model

OpenAI has announced the release of ChatGPT Images 2.5. According to Simon Willison, who tested the novelty, it brings two new models to the API: gpt-image-2.5-sunburst and gpt-image-2.5-flare. OpenAI states that over 3 billion images have already been generated across ChatGPT Images and API models.

Sunburst targets precision and edits

While Sunburst focuses on precision for more complex edits, Flare is designed for fast generation of everyday images.

Support for reference images in CLI tools is growing

Willison has updated his CLI tool openai_image.py, which now supports inputting reference images. He tested it on a chart of internal coding agent adoption, successfully adding an illustration of an examining raccoon in a lab coat using the Sunburst model.

Accuracy limits for abstract concepts

Even though Sunburst can add a raccoon, it still struggles with accurately rendering complex business abstractions or text if not provided with very tight context.

What will decide adoption

The real test will be the API's stability under load and Sunburst's ability to maintain accuracy in production deployments for enterprise clients.

Lilith's verdict

The Sunburst model is the first interesting thing since DALL-E 3, as it can finally be used as a reliable graphic tool.