ChatGPT Images 2.5: Faster Generation, More Precise Edits, and Two New API Models

ChatGPT Images 2.5 is OpenAI’s new image-generation and editing system, announced September 8, 2026. The update is designed to make the creative loop faster and more dependable: generate an image, request a targeted change, and continue refining it without losing the elements that were already working. OpenAI says it reduces generation latency by up to 50% versus Images 2.0. 9

What changed from Images 2.0

The headline improvement is speed, but the update is also about fidelity and edit control. According to OpenAI, Images 2.5 produces sharper details, more natural lighting, and richer textures. It is also intended to preserve subjects from reference photos more effectively and follow editing instructions more reliably over multiple turns. 9

That matters when an image needs refinement rather than a complete reroll. Instead of recreating the entire scene for each request, the goal is to modify the requested element while retaining the subject, composition, and visual choices a user wants to keep.

OpenAI also positions the model for more structured visual work, including complex layouts, transparent-background assets, and information-rich graphics. Those uses still benefit from human review: improved instruction following is not the same as a guarantee that text, facts, or visual relationships in an infographic are correct. 9

Better control over edits and composition

Images 2.5 is built for requests that change one aspect of an image while preserving the rest. Examples include:

• placing a reference subject in a different setting;
• changing an image’s style or composition;
• making focused adjustments across several rounds of feedback; and
• creating more structured design assets rather than only open-ended illustrations. 9

For creators, the useful distinction is between generating a new concept and art-directing an existing one. The latter is where better subject preservation and multi-turn instruction following should have the most visible impact.

New creation tools in ChatGPT

OpenAI paired the model update with several workflow features:

• Sketch: Draw directly in ChatGPT and use the drawing as a visual reference for generation.
• Templates: Start from common formats, including items such as flyers and product photos.
• Image comments: Add feedback directly to an image for more focused revisions.
• Prompt sharing: Share reusable prompts that others can adapt. 9

Together, these tools make it easier to begin with a rough visual direction and then turn it into a more polished asset through iteration.

API models: Flare vs. Sunburst

ChatGPT Images 2.5 is the product name inside ChatGPT. For developers, OpenAI introduced two separate API models rather than one generic gpt-image-2.5 endpoint. 9

gpt-image-2.5-flare

Flare is the general-purpose option. OpenAI says it brings the same quality, editing, and speed improvements as Images 2.5, making it the logical choice when responsiveness and repeated generation are priorities. 9

Choose Flare for applications such as rapid prototyping, high-volume image workflows, and product experiences where users benefit from shorter waits.

gpt-image-2.5-sunburst

Sunburst is aimed at detailed creative work where editing precision is worth a longer generation time. 9

Choose Sunburst when a result needs especially careful control—for example, refined campaign creative, product imagery, or edits that must hold together across multiple rounds.

Availability and subscription coverage

OpenAI said ChatGPT Images 2.5 was rolling out to ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web. Its announcement describes the rollout as reaching all users of those products. 9

The available announcement material does not provide a complete plan-by-plan breakdown of image limits, regional differences, or the precise timing of access for every account. In practice, a rollout announcement should not be read as a promise that every user will see identical limits immediately.

Launch-day reliability context

On September 8, OpenAI’s status page recorded elevated errors affecting image generation in both ChatGPT and the Images API. The incident was later marked resolved. 19

A separate status incident around the same period concerned elevated errors for ChatGPT file uploads and Codex Cloud task creation. 23 These incidents provide useful operational context for users who encountered failures during the rollout, but the published status updates do not establish that Images 2.5 caused them.

The bottom line

ChatGPT Images 2.5 is primarily an iteration upgrade: faster image generation, more natural-looking output, and better control when refining an image over several edits. For everyday API use, Flare is the speed-oriented default; Sunburst is the slower option for precision-focused creative work. 9

The most meaningful test is not a single prompt. It is whether the system can preserve the important parts of an image while accurately carrying out a sequence of specific changes—and whether the final asset still holds up under human review.