OpenAI releases GPT Image 2.5, adding Sunburst and Flare models

OpenAI shipped GPT Image 2.5 on September 8, 2026, replacing GPT Image 2 as the default image-generation model across ChatGPT and the API with two new variants: GPT Image 2.5 Flare and GPT Image 2.5 Sunburst.

What's new

The headline change is speed and fidelity. OpenAI says it has reduced image generation latency by up to 50% compared with the prior generation, while also improving lighting realism, texture detail, and adherence to multi-turn editing instructions.

The release ships as two distinct API models rather than a single upgrade:

• GPT Image 2.5 Flare is the new default for the Image API and the Responses API image-generation tool. It targets everyday, high-volume use cases — social content, product imagery, visual search, and rapid prototyping — at GPT Image 2 token rates.
• GPT Image 2.5 Sunburst is aimed at premium, editing-heavy workflows that need tighter control over specific elements of an image, such as production-ready campaign creative or polished product photography, at the cost of longer generation times.

Both models add new xhigh and max quality settings on top of the existing tiers, and both are reachable through the same Image API and Responses API surfaces GPT Image 2 already used. On the consumer side, the base Images 2.5 model is rolling out to ChatGPT, ChatGPT Work, and Codex across desktop, mobile, and web.

OpenAI also highlights better reference-image handling: when a user provides a photo of a real subject, the company says 2.5 preserves more of that subject's distinctive features when placing it in new settings, styles, or compositions — an area where GPT Image 2 was prone to drifting off-likeness across edits.

Context

GPT Image 2 launched as OpenAI's prior-generation image model and has been the backbone of ChatGPT's built-in image generation as well as third-party API integrations. OpenAI notes that more than 3 billion images are now created weekly across ChatGPT Images and the GPT-Image API family combined, underscoring why the company is optimizing for both raw throughput (Flare) and precision editing (Sunburst) rather than shipping one general-purpose successor.

The two-model split mirrors a pattern OpenAI has used elsewhere in its lineup — pairing a fast, cost-efficient default with a slower, higher-fidelity variant for professional workflows — rather than forcing every use case through a single set of latency/quality tradeoffs.

Why it matters

Image generation has become one of the highest-volume, most price-sensitive workloads in the API, and a 50% latency cut directly lowers the cost of serving interactive, high-throughput products such as social apps, ad-creative tools, and e-commerce visual search. Splitting the lineup into Flare and Sunburst also signals that OpenAI sees enough demand at both ends — cheap-and-fast versus precise-and-controllable — to maintain two supported models rather than one compromise. For developers already on GPT Image 2, the same token pricing on Flare removes cost as a reason to delay upgrading.