GPT Image 2.5: Flare vs Sunburst and When to Upgrade

Update, September 9, 2026: GPT Image 2.5 is live on WaveSpeedAI. The practical sections below describe the four endpoints and their current generation controls.

Short version: choose the endpoint that matches the workload: Flare Text-to-Image, Flare Edit, Sunburst Text-to-Image, or Sunburst Edit.

What is available now

The GPT Image 2.5 API overview brings the four WaveSpeedAI endpoints together in one place.

GPT Image 2.5 arrives as two tiers, each with a text-to-image and an edit endpoint:

• Flare Text-to-Image and Flare Edit: the fast, balanced tier for everyday generation at low latency.
• Sunburst Text-to-Image and Sunburst Edit: the precision-focused tier that spends more time per image for extra fidelity on intricate detail.

The corresponding endpoints expose the same core generation controls: prompt, 15 aspect_ratio presets from 1:3 to 3:1, resolution at 1k, 2k, or 4k, five quality levels (low, medium, high, xhigh, max), and output_format of png, jpeg, or webp. Edit also requires images and supports up to 16 reference images. For request examples and parameter defaults, see the GPT Image 2.5 API guide.

Flare vs Sunburst: which should you choose?

Use the following as a starting point for evaluation, rather than a benchmark ranking. Compare both models with the same prompt, reference images, resolution, and quality setting.

Your workload	Start with	What to inspect
Exploring layouts or campaign concepts	Flare Text-to-Image	Prompt adherence, usable variations, and turnaround time
Repeated background or styling edits	Flare Edit	Whether the subject and its proportions remain consistent
Intricate product or architectural imagery	Sunburst Text-to-Image	Fine detail, material texture, and composition
Edits where preserving specific details matters	Sunburst Edit	Unrequested changes to reference details and accuracy of the requested edit

For text-heavy designs, test the exact wording and inspect the output at its intended display size. A model’s positioning alone does not establish that it will handle your typography or product details better.

How the quality tiers relate to GPT Image 2

GPT Image 2.5 uses a five-step quality scale: low, medium, high, xhigh, and max. The tiers are not defined as fidelity equivalents of similarly named tiers in older models, so start at medium, compare against your existing outputs on the same prompts, and adjust only where a specific detail is missing.

What this means if you are building an image feature

1. Migrate by model ID, then validate the payload. Choose the appropriate Flare or Sunburst text-to-image/edit ID. Confirm the current parameter names, allowed values, and defaults on that model page before copying production traffic.

2. Run your evaluation set. If you saved the prompts that gave GPT Image 2 trouble, small text, dark scenes, skin under hard light, multi-step edits, run them through both tiers now. That same-prompt comparison is worth more than any leaderboard screenshot.

3. Draft on Flare, evaluate on Sunburst. Iterate on Flare when speed matters, then test the settled prompt on Sunburst when the deliverable needs extra precision.

4. Keep the existing endpoint in place until you are satisfied. Do not force a cutover before the new endpoint passes your quality and latency checks.

FAQ

Is GPT Image 2.5 available through an API?

Yes. Since September 9, 2026 WaveSpeedAI serves four GPT Image 2.5 endpoints: openai/gpt-image-2.5-flare/text-to-image, openai/gpt-image-2.5-flare/edit, openai/gpt-image-2.5-sunburst/text-to-image, and openai/gpt-image-2.5-sunburst/edit.

What is the difference between Flare and Sunburst?

Flare is the fast, balanced tier for everyday generation at low latency. Sunburst is the precision-focused tier for extra fidelity on intricate detail. Their corresponding endpoints expose the same core generation controls.

Do I have to change my GPT Image 2 integration?

Change the model ID, then validate the current parameter names, allowed values, and defaults on the selected GPT Image 2.5 model page. Roll out one endpoint at a time and keep the existing endpoint as a fallback until testing is complete.