GPT Image 2.6 vs 2.5: What Changes for Sunburst and Flare?

GPT Image 2.6 keeps the two-model choice introduced with 2.5: Flare for faster iteration and Sunburst for demanding visual work. The useful comparison is therefore two upgrades: 2.5 Flare to 2.6 Flare, and 2.5 Sunburst to 2.6 Sunburst.

OpenAI announced Images 2.6 on November 4, 2026. It adds faster generation, improved editing consistency, animated GIFs, SVG output, and a new control over scene planning. Here is how those changes affect the choice between the four model IDs.

GPT Image 2.6 vs 2.5: Four Model IDs, Two Familiar Roles

The names repeat across generations, but each version has its own API identifier. Sunburst and Flare are separate model choices within each release.

OpenAI's 2.5 launch announcement positioned Flare as the default for most applications, including rapid prototyping and high-volume generation. Sunburst targeted premium creative workflows with tighter control across edits. The 2.6 Flare model page and 2.6 Sunburst model page retain that practical division.

What Changes for Flare and Sunburst?

Flare: Faster Feedback and Better Layouts

OpenAI reports that Images 2.6 generates images 50% faster than before. More specifically, 2.6 Flare can generate an image in as little as six seconds at low quality. OpenAI also reports improved Flare layouts for posters, product imagery, and other professional visuals.

For a creator exploring several directions, that makes Flare the natural starting point. The six-second figure is a best-case example under a stated quality setting; it should not become a promise for every request.

Sunburst: Detail and Continuity Across Edits

For 2.6 Sunburst, OpenAI highlights higher-quality generation and stronger consistency across edits at medium or high quality. That is relevant when a product, character, or brand detail must remain recognizable through revisions.

The 2.6 system card also describes improved detail preservation and responses to sketches and image annotations across the updated models. These are useful improvements for an iterative workflow, although they do not establish perfect preservation on every edit.

New Controls and Output Formats

Complexity Separates Planning From Visual Quality

The new complexity parameter controls how much the model plans before rendering. quality controls visual detail. OpenAI's image prompting documentation lists low, medium, and high complexity, with high used when the field is omitted.

That distinction matters: a small background adjustment and a crowded poster have different planning needs. Compare complexity settings while keeping quality fixed, then adjust quality separately to judge lettering or texture.

GIFs and SVGs Expand the Deliverables

Both 2.6 models support animated GIF and SVG output through the Image API. GIFs add short animated assets; SVGs provide vector paths, shapes, and colors useful for editable graphics.

The image generation guide documents an important boundary: these formats use the image generation endpoint, and the image edit endpoint does not output GIFs or SVGs. SVG generation also involves a reference image followed by an authoring step.

Markup Editing Also Reaches Supported 2.5 Models

Markup lets an annotation layer indicate where an edit should happen. The November 4 API changelog explicitly says markup-guided editing was added for supported GPT Image 2.5 and 2.6 models. It is part of this release's API update, so it should not be presented as an exclusive reason to move from 2.5 to 2.6.

GPT Image 2.6 Performance: What the Official Evidence Shows

The launch supplies a headline speed improvement and a conditional Flare generation time. It does not provide a four-model visual-quality benchmark table with identical prompts, dimensions, and quality settings. Those figures alone cannot establish which variant wins every task.

OpenAI's system card evaluates safety behavior, including safeguards for GIF creation. Those evaluations answer different questions from image quality or editing accuracy. They should not be repurposed as a creative-performance ranking.

GPT Image 2.6 vs 2.5 API Pricing

As of November 5, 2026, OpenAI's Standard API pricing lists the same token rates for all four models:

These are token prices, not flat prices per image. Actual consumption can vary by model, size, and quality. Flare's speed positioning does not imply a lower published token rate.

Cached rates apply only to eligible cached inputs; the guide limits 2.5 caching to the Responses API image tool. SVG output adds authoring-model token charges to reference-image costs. The 2.5 Flare and 2.5 Sunburst pages separately list Batch image output at $15 per million tokens; that is distinct from Standard pricing and should not be assumed for 2.6.

Which Model Should You Test First?

For GPT Image 2.6 vs 2.5, begin with the corresponding successor: Flare to Flare, or Sunburst to Sunburst. OpenAI recommends keeping prompts, reference images, dimensions, and quality consistent for the initial comparison.

Start with 2.6 Flare when fast feedback matters most. Start with 2.6 Sunburst when precise details determine whether an image is usable. Once Sunburst meets the brief, try Flare on the same task. The decision should follow acceptable results, response time, and measured cost for your actual work.