ChatGPT Images 2.5 Launches With Stronger Reference Preservation and Two API Models

ChatGPT Images 2.5 launches with stronger reference preservation and iterative editing, alongside separate GPT-Image-2.5 Flare and Sunburst API models. Flare prioritizes speed and volume, while Sunburst trades longer generation times for tighter precision.

• ChatGPT gets an updated image experience. ChatGPT Images 2.5 is available across desktop, mobile, and web.
• API customers get two distinct models. OpenAI positions Flare as the faster default and Sunburst as the precision-focused option.
• Iterative editing gains more control. OpenAI claims improvements to reference-subject preservation, targeted editing, and multi-turn editing.
• Generation latency can fall by up to 50%. OpenAI compares that figure with Images 2.0 but does not provide separate latency numbers for Flare and Sunburst.

ChatGPT Images 2.5 targets more consistent references and successive edits

OpenAI says ChatGPT Images 2.5 improves reference-subject preservation, targeted editing, and multi-turn editing. The update focuses on retaining established subjects and details as instructions change across successive passes.

The company also claims improvements to lighting, textures, complex layouts, and transparent backgrounds. Those additions broaden the types of image elements that can be created or modified within the ChatGPT experience across desktop, mobile, and web.

The release follows GPT-Image-2’s appearance on Chatbot Arena, where early examples previewed OpenAI’s next image-generation release. OpenAI claims Images 2.5 can reduce generation latency by up to 50% compared with Images 2.0, although the supplied materials do not break that figure down by product or API model.

Flare prioritizes speed, prototyping, and higher-volume generation

OpenAI positions GPT-Image-2.5 Flare as the faster default API model for creator and social content, prototyping, and volume work. Its stated role favors workloads where teams need faster generation or more outputs rather than the tighter precision assigned to Sunburst.

That positioning makes Flare the documented choice for rapid iteration and higher-volume production. The official announcement does not provide a separate latency figure for Flare or state that every broader Images 2.5 improvement performs identically across both API models.

Sunburst prioritizes tighter precision despite longer generation times

OpenAI positions GPT-Image-2.5 Sunburst for premium creative and editing workflows where tighter precision matters. The company says Sunburst takes longer to generate images than Flare, establishing a direct API choice between faster output and closer control.

Sunburst therefore fits the precision side of OpenAI’s documented model split, while Flare covers faster prototyping and volume work. OpenAI’s product description does not say that projects or edits transfer directly between the models, so the source supports treating them as separate API choices rather than a documented two-stage workflow.