OpenAI released GPT Image 2.5 on September 8, 2026, as two models rather than one: GPT Image 2.5 Sunburst and GPT Image 2.5 Flare. Both handle image generation and editing through the Image API and the image generation tool in the Responses API. OpenAI's guidance is to use Sunburst "for workflows where editing precision matters most" and Flare "for fast, high-quality everyday image generation."
The Sunburst model page calls it OpenAI's most capable model for image generation and editing. It accepts text and image inputs, supports inpainting, and adds two new quality levels, xhigh and max, on top of low, medium, high and auto. The default snapshot is gpt-image-2.5-sunburst-2026-09-08. Pricing is listed at $8 per million image input tokens ($2 cached), $30 per million image output tokens and $5 per million text input tokens; OpenAI's changelog says both variants use GPT Image 2 token rates.
Splitting a release into a precise, slower tier and a fast, everyday tier mirrors how text model families are now sold, and it signals that image generation has moved from novelty to production workload where latency and cost per image matter as much as peak quality. The new quality levels let developers buy more compute per image when the result will be printed, cropped or scrutinized.
The claims about quality are OpenAI's own; the developer documentation fetched for this entry gives no benchmark numbers or independent evaluations, and speed comparisons between the two variants circulating elsewhere were not confirmed in OpenAI's pages. Whether the higher quality settings are worth their cost is something users will have to measure on their own images.