OpenAI split the 2.5 release into gpt-image-2.5-flare, the default choice for most applications, and gpt-image-2.5-sunburst, built for premium visual workflows where editing precision matters most. Flare is the everyday pick; Sunburst trades longer generation time for tighter control across edits. Both sit in the Sunra model picker next to GPT Image 2, so switching is one click, not a new account.
OpenAI reports that Images 2.5 cuts image generation latency by up to 50% compared with Images 2.0, and that Flare delivers higher-quality images than GPT Image 2 at 50% lower latency. Faster turns mean more takes per session — the difference between exploring five directions and settling for the first usable one.
2.5 is better at preserving the subjects in your reference photos. Attach up to 16 reference images and the face, product, or brand element you started with survives the generation — the failure mode where an edit returns 'close but not the same person' is exactly what this release targets.
The model follows editing instructions more reliably across multiple turns: it changes the requested elements and leaves the rest of the image intact, and earlier edits carry through longer conversations without degrading. Sunburst is the variant OpenAI points at for this kind of iterative, precision work.
OpenAI describes more natural lighting and richer textures, more accurate rendering of real-world information, consistent visual styles across a set, and complex layouts including transparent backgrounds. For packaging, signage, and UI mockups that means fewer post-processing passes before the asset is usable — and the text-in-image editor still covers the last-mile copy fixes.
Both 2.5 models are listed by OpenAI at the same rates as GPT Image 2: text input $5.00, image input $8.00, and image output $30.00 per 1M tokens (cached input $1.25 / $2.00). 2.5 is a quality and speed upgrade, not a price increase. On Sunra you pay in credits per image — the price depends on aspect ratio, resolution (1K–4K), and quality tier, and is shown before you generate; see the pricing page.
Go to Sunra Image and select GPT Image 2.5 Flare from the model picker. Free daily credits are available immediately; no OpenAI account or API key needed.
Describe the scene, the subject, and any exact colors (hex values work). Attach up to 16 reference images when you need a specific face, product, or style preserved — add one and the image-editing endpoint is chosen for you.
Pick an aspect ratio (3:2, 1:1, 2:3, 16:9, 9:16, 4:3, or 3:4), a resolution of 1K, 2K, or 4K, and one of five quality tiers — low, medium, high, xhigh, or max. Higher tiers add detail — and latency and credits — so start at high and step up to xhigh or max only for the final render.
When a single element has to change and everything else has to stay, select GPT Image 2.5 Sunburst and iterate — earlier edits carry through the session. Want motion afterwards? Turn the image into a video clip without leaving the platform.
OpenAI positions Flare for creator and social content, product experiences, and rapid image prototyping — the high-frequency, many-variations work where 50% lower latency compounds. Generate a batch, pick the keeper, post.
Sunburst is built for production-ready campaign creative and polished product imagery where every edit has to land precisely. Upload the hero product shot, iterate backgrounds and props across several turns, and keep the product itself pixel-faithful throughout.
Flare's brief also names visual search and high-volume generation: catalog variants, thumbnail sets, localized versions of one master asset. Pair with your project to run the same prompt across GPT Image 2.5 and other models side by side.
Because earlier edits now carry through longer conversations without degradation, multi-step retouching — swap the sky, then the wardrobe, then the copy on the sign — stays coherent from the first turn to the last instead of slowly losing the original.
Join thousands of creators, agencies, and brands who use Sunra every day.
