GPT Image 2.5(sunburst) API

GPT-Image-2.5 Specifications

The two variants share much of the same API surface, but their optimization targets are different.

What Is GPT-Image-2.5 Sunburst?

GPT-Image-2.5 Sunburst is OpenAI's premium image-generation and editing model in the GPT Image 2.5 family. It is designed for visual workflows where precise editing, detailed creative control, and consistency across edits are more important than minimizing generation time.

OpenAI introduced Sunburst alongside GPT-Image-2.5 Flare on September 8, 2026. While Flare is positioned as the faster choice for high-quality everyday image generation, Sunburst is designed for premium visual workflows that benefit from tighter control across edits and is intended for production-ready campaign creative and polished product imagery.

The distinction is therefore primarily about workflow requirements rather than simply image quality. Sunburst is the better fit when an application needs to make controlled changes to an existing image, preserve important visual characteristics, or refine a creative asset through multiple editing passes.

Key Features of GPT-Image-2.5 Sunburst

Precision-Focused Image Editing

The defining characteristic of Sunburst is its emphasis on editing precision. OpenAI specifically recommends it for workflows where editing precision matters most. This makes it appropriate for production workflows in which changing one part of an image while preserving the rest is more important than simply generating an acceptable image quickly.

Text-to-Image Generation

Sunburst can generate images from natural-language prompts. Developers can describe the desired subject, composition, visual style, environment, lighting, and other creative requirements and use the resulting image as a production asset or starting point for additional edits.

Image-to-Image Editing

Sunburst accepts image inputs, enabling developers to build editing workflows around existing images rather than generating every asset from scratch.

This is particularly useful when an application needs to modify an existing product photograph, campaign asset, character, scene, or other visual while maintaining important elements of the source image.

Improved Consistency Across Edits

OpenAI's GPT Images 2.5 announcement highlights better preservation of subjects from reference photos and more reliable instruction following across multiple editing turns. These improvements are particularly relevant to Sunburst because its intended use cases involve detailed, iterative creative workflows.

Flexible Quality Settings

Sunburst supports low, medium, high, xhigh, max, and auto quality settings. This gives developers control over the quality-versus-resource trade-off depending on whether they are generating an initial concept, previewing an edit, or producing a final asset.

Benchmark Performance of GPT-Image-2.5 Sunburst

One reported Arena-based evaluation placed GPT-Image-2.5 Sunburst at 1,421 Elo for text-to-image generation, ahead of GPT-Image-2.5 Flare at 1,399 and GPT Image 2 at 1,381. The same source reported Sunburst at 1,520 on an image-editing leaderboard, ahead of Flare at 1,491 and GPT Image 2 at 1,461. These figures should be treated as early preference-test results rather than universal image-quality scores, because Arena scores depend on the evaluation population, task mix, model version, and testing period.

Third-Party Benchmark Results

*Tosea's published table gives cumulative drift after each editing round; the source reports three-round cumulative drift of 9.9% for Sunburst and 9.2% for Flare, versus 11.4% for GPT Image 2.

Independent Multi-Turn Editing Test

A separate production-oriented test by Tosea evaluated Sunburst, Flare, and GPT Image 2 using a five-page 16:9 presentation workflow with reference images and repeated editing instructions. The test used 41 successful API calls across the three models.

In its three-round editing experiment, Tosea measured the percentage of pixels that changed beyond a perceptual threshold outside the requested modification. Sunburst showed 4.1% drift after the first round, 6.8% after the second, and 9.9% after the third, compared with 5.6%, 8.1%, and 11.4% for GPT Image 2.

That suggests an important practical advantage: Sunburst may preserve more of an existing composition during repeated edits than GPT Image 2.

GPT-Image-2.5 Sunburst vs. GPT-Image-2.5 Flare

Sunburst and Flare are the two API models introduced as part of GPT Image 2.5, but they target different production requirements.

OpenAI explicitly recommends Flare as the default choice for most applications, while Sunburst is intended for premium visual workflows requiring tighter control across edits.

When Should You Choose Sunburst?

Choose Sunburst when the output is intended to be a polished production asset and small editing errors can create significant downstream work.

Examples include a product photograph where the product must remain visually consistent, a campaign image that needs several controlled revisions, or a creative asset that must preserve a subject across multiple editing rounds.

When Should You Choose Flare?

Flare is generally more appropriate when the application generates many images, requires rapid previews, or prioritizes low latency.

A useful rule is simple: choose Sunburst for precision; choose Flare for speed and iteration volume.

GPT-Image-2.5 Sunburst Use Cases

Production-Ready Campaign Creative

Marketing teams can use Sunburst to develop polished advertising visuals where composition, subject identity, typography, and other visual details need to remain consistent through multiple revisions.

Product Photography

Sunburst is particularly well suited to product imagery that begins with an existing reference image. Developers can build workflows that modify backgrounds, environments, compositions, or other visual elements while keeping the underlying product recognizable.

Brand and Creative Asset Production

Brands can use the model to create and refine visual assets while maintaining consistent subjects and design characteristics across successive generations and edits.

Professional Image Editing

Applications that expose AI editing as a core feature can use Sunburst when users need controlled transformations rather than completely new images.

Multi-Turn Creative Workflows

Sunburst is well suited to workflows where an image is generated first and then progressively refined through multiple instructions. This aligns with OpenAI's emphasis on improved instruction following and subject preservation across multiple edits.

GPT-Image-2.5 Sunburst Limitations

Sunburst should not be treated as a general-purpose multimodal language model. Its documented input and output modalities center on text and images, while audio and video are unsupported.

The model also does not currently support streaming, function calling, structured outputs, or fine-tuning according to OpenAI's model specification.

Another practical consideration is latency. Sunburst is deliberately positioned for precision-oriented workflows and can take longer to generate than Flare. Applications where users expect extremely rapid image iteration may therefore benefit from Flare instead.

How to Access GPT-Image-2.5 Sunburst API with CometAPI

Step 1: Get a CometAPI API Key

Create or sign in to your CometAPI account and obtain an API key from the CometAPI console.

Step 2: Select GPT-Image-2.5 Sunburst

Use the model ID:

gpt-image-2.5-sunburst

For a generation workflow, provide a text prompt describing the desired image. For an editing workflow, provide the required image input together with the instructions describing what should be changed.

Step 3: Configure the Image Generation Request

Select the quality level appropriate for the task. Lower settings can be useful during experimentation, while higher settings are more appropriate for final creative assets.

Because Sunburst is specifically intended for precision-oriented work, production applications should test the model with representative reference images and editing instructions before committing to a particular quality configuration.

Step 4: Process the Generated Image

After the API request completes, process the returned image according to the response format provided by CometAPI. Production applications should also implement appropriate error handling, retry logic, storage, and output validation.

GPT-Image-2.5 Sunburst vs. GPT Image 2

GPT-Image-2.5 Sunburst is a newer model in OpenAI's GPT Image 2.5 generation. OpenAI positions the GPT Image 2.5 models as bringing sharper details, more natural lighting, richer textures, better preservation of reference subjects, and more reliable editing instructions.

For developers migrating from GPT Image 2, Sunburst is particularly relevant when the application has moved beyond basic image generation and requires more controlled editing and production-quality creative workflows.

Who Should Use GPT-Image-2.5 Sunburst?

GPT-Image-2.5 Sunburst is best suited to developers building professional image-generation and editing applications where output quality and editing precision have a higher priority than minimum latency.

It is particularly relevant to:

• Advertising and campaign-generation platforms
• Ecommerce product-image systems
• Professional creative tools
• Brand-content generation
• AI image-editing applications
• Product visualization workflows
• Multi-step image refinement systems

For applications that primarily need fast image generation at high volume, GPT-Image-2.5 Flare is likely the better starting point. For applications where users repeatedly refine a visual and expect the model to preserve important details, Sunburst is the more appropriate choice.