GPT Image 2.5: Flare vs Sunburst, pricing and features
GPT Image 2.5 is OpenAI's image-generation and editing model family, released on September 8, 2026. It comes in two variants: Flare for fast, everyday creation and Sunburst for detailed work that needs more precise edits. Both can generate images from text, work with reference images, and produce transparent backgrounds.
The unusual part is the pricing: Flare and Sunburst share the same token rates as GPT Image 2. Choosing between them is about speed, precision, and actual usage, not a higher price for each token.
Pricing and availability checked September 9, 2026. Provider support can change.
GPT Image 2.5 Flare vs Sunburst
| Comparison | Flare | Sunburst |
|---|---|---|
| Best starting point for | Exploring ideas and everyday image generation | Precise edits and fine-detail creative work |
| Speed and precision | Faster iteration | Greater editing precision, with longer generation times |
| Example workflows | Concept drafts, social images, visual exploration | Product imagery, campaign assets, controlled revisions |
| OpenAI API model ID | gpt-image-2.5-flare | gpt-image-2.5-sunburst |
| Token prices | Same rate card | Same rate card |
Both accept text and image inputs and produce images. Each offers five
fixed quality settings—low, medium, high, xhigh, and max—plus
auto. The variant and the quality setting are separate choices:
Sunburst is not simply Flare with quality turned up. See OpenAI's
Flare model card
and Sunburst model card
for the current model specifications.
What changed from GPT Image 2?
OpenAI's Images 2.5 announcement describes improvements in four areas:
- Image fidelity: more natural lighting and texture, with better preservation of subjects from reference photos.
- Targeted editing: change the requested part of an image while keeping the surrounding composition and details.
- Successive revisions: carry earlier edits forward more consistently across a sequence of changes.
- Speed: OpenAI reports up to 50% lower generation latency compared with Images 2.0. This is OpenAI's claim, not a Grida benchmark.
For example, a product-image workflow might keep the same bottle and composition while changing the background, then refine the lighting in a second edit. This is the kind of controlled revision 2.5 is designed to improve; it is not a guarantee that every generated detail will be correct.
GPT Image 2.5 pricing compared with GPT Image 2
The table shows OpenAI's standard API rates in USD per million tokens, not a fixed price per image or a discounted Batch rate.
| Token type | GPT Image 2 | 2.5 Flare | 2.5 Sunburst |
|---|---|---|---|
| Text input | $5.00 | $5.00 | $5.00 |
| Cached text input | $1.25 | $1.25 | $1.25 |
| Image input | $8.00 | $8.00 | $8.00 |
| Cached image input | $2.00 | $2.00 | $2.00 |
| Image output | $30.00 | $30.00 | $30.00 |
Sources: the GPT Image 2 model card, the Flare and Sunburst model cards above, and OpenAI API pricing.
How much does one image cost?
Tokens are the units the API meters, not a count of pictures. The final cost depends on the tokens consumed by the prompt, reference images, and generated output. Model choice, quality, and dimensions can change that usage even when the rate card is identical.
For a size-and-quality estimate, use the model selector in OpenAI's image-generation cost calculator and select GPT Image 2.5. It groups Flare and Sunburst together and gives these examples for a 1024 × 1024 image:
| Quality | Estimated image-output tokens | Image-output cost |
|---|---|---|
medium | 439 | $0.01317 |
high | 1,756 | $0.05268 |
These estimates exclude text and image inputs and any streaming partial images; they are not total request prices. Do not reuse GPT Image 2's per-image table for 2.5. Compare the provider's reported usage for your own workflow; neither variant is guaranteed to cost less on every request.
Does GPT Image 2.5 support transparent backgrounds?
Yes. Both variants support native transparent-background generation.
In OpenAI's API, request background: "transparent" with PNG or WebP
output. JPEG does not preserve transparency. See the
OpenAI output settings guide.
Native transparency creates an image with an alpha channel, which can be placed over other artwork. It is different from generating a solid-color background and removing it afterward. For stickers, isolated product assets, or overlays, check that both the selected provider and the output format expose this capability.
Using GPT Image 2.5 in Grida
Both variants are available in Grida using purchased AI credits. Text-to-image generation does not require your own provider API key. You can also connect Vercel AI Gateway, fal, or OpenRouter keys. Native model support and the options available in a particular application are not always the same:
| Access in Grida | Text-to-image | Reference-image editing | Transparent output |
|---|---|---|---|
| Grida credits | Yes | No | Yes |
| Vercel AI Gateway key | Yes | No | Yes |
| fal key | Yes | Yes | Yes |
| OpenRouter key | Yes | Yes | Not enabled in Grida |
Reference editing requires an agent integration that explicitly selects one of these variants and uses a connected fal or OpenRouter key. The desktop image playground generates from text only. Desktop users need version 0.0.22 or later for these models. For provider-specific prices, settings, and billing details, see Models & Pricing.
Is GPT Image 2 being removed?
No. Grida marks GPT Image 2 as deprecated, meaning it is a legacy choice, but keeps it listed and callable. Existing defaults and saved model selections are unchanged. This is a Grida catalog decision, not an announcement that OpenAI has retired GPT Image 2.