GPT Image 2.5 – Flare and Sunburst on one endpoint
GPT Image 2.5 comes as two OpenAI models that take the same request: draft on Flare, then switch the ID to Sunburst for the final edit. One 1024×1024 image at low quality costs about $0.021 on the Standard route, and a failed render costs nothing.
- Text to image
- Edits with up to 16 references
- Masked edits
- Transparent PNG
- Up to 3840 px per side
- Six quality levels
Run GPT Image 2.5 here first
Pick GPT Image 2.5 Flare or Sunburst and a billing route, then generate or edit. The form sends the same fields as the API.
gpt-image-2.5-flare$0.021 /image
Generate from a prompt or edit reference images. Submit once; results appear when the task completes.
required · up to 32,000 characters
Reference images for image-to-image and editing. Files are uploaded and sent as URLs; on Official IDs each one adds image input tokens.
Let the model choose the image dimensions.
auto or WIDTHxHEIGHT · custom sizes supported
Reference images for image-to-image and editing. Files are uploaded and sent as URLs; on Official IDs each one adds image input tokens.
PNG smaller than 4 MB, matching the first reference image dimensions. Transparent areas are edited.
default auto · xhigh and max are new
1 to 10 · default 1
auto, opaque or transparent · transparency requires PNG
default png
Choose JPEG to set compression.
default auto
Greyed-out options relax content filtering. They are available through the API only, not in the playground.
Standard estimates use the current per-image rate for the selected model and quality. Official estimates use a 1024×1024 sample at the selected quality. Actual size, prompt and reference images can change an Official charge.
Try one of these prompts
Rendered with GPT Image 2.5 on TokLab
GPT Image 2.5 pricing
Every GPT Image 2.5 price here is read from the rates that bill your account, at the moment the page loads.
GPT Image 2.5 cost at each quality level
One 1024×1024 image per row. Standard prices are per image; Official prices multiply our measured tokens by the live token rate.
| Category | gpt-image-2.5-flarePrice · 1 image | gpt-image-2.5-sunburstPrice · 1 image | gpt-image-2.5-flare-officialPrice · 1 image | gpt-image-2.5-sunburst-officialPrice · 1 image |
|---|---|---|---|---|
| low | $0.021 | $0.021 | $0.00448 | $0.00448 |
| medium | $0.021 | $0.021 | $0.00995 | $0.00995 |
| high | $0.021 | $0.021 | $0.03958 | $0.03958 |
| xhigh | $0.0225 | $0.0225 | $0.07031 | $0.07031 |
| max | $0.0225 | $0.0225 | $0.15811 | $0.15811 |
| auto (model picks) | $0.021 | $0.021 | $0.00522 | $0.00522 |
GPT Image 2.5 output tokens we recorded at 1024×1024: low 196, medium 439, high 1,756, xhigh 3,122, max 7,024. Other sizes and prompts give other counts.
Estimate a GPT Image 2.5 request
Only the token-billed Official IDs need an estimate. Set quality, size and count; the figure uses the live rate.
This is an estimate made before you submit. Your usage history shows what each task actually cost.
gpt-image-2.5-flare-official: Image output tokens $22.50 / 1M · Text input tokens $3.75 / 1M
gpt-image-2.5-sunburst-official: Image output tokens $22.50 / 1M · Text input tokens $3.75 / 1M
About GPT Image 2.5
GPT Image 2.5 is OpenAI's image model, released on September 8, 2026 as two models. Flare is the faster everyday model. Sunburst is slower and keeps more of the original picture when you edit.
Compared with GPT Image 2, GPT Image 2.5 adds two quality steps above high: xhigh and max. On TokLab each model has a Standard route billed per image and an Official route billed per token.
Endpoint POST /v1/images/generations · Results via GET /v1/tasks/{id} · Up to 16 reference images · Mask edits on the first reference · n from 1 to 10 · URL inputs only
Call GPT Image 2.5 from code or an agent
Both paths use the same endpoint and the same balance.
GPT Image 2.5 from your backend
Send JSON, keep the task ID, poll until it completes. Field names follow the OpenAI Images API, with URLs instead of uploads.
- 1Create an API key
- 2POST /v1/images/generations
- 3Poll GET /v1/tasks/{id}
- 4Copy output.data[].url to your storage
GPT Image 2.5 in Claude Code or Codex
Paste the prompt into your agent. It reads TOKLAB_API_KEY, shows you the request and its cost, and waits for your approval.
- 1Export TOKLAB_API_KEY in your shell
- 2Copy the prompt
- 3Fill in subject, size and quality
- 4Approve the request it shows you
How a GPT Image 2.5 request works
One endpoint for generating and editing. Send JSON, get a task ID, collect the images when they are ready.
OpenAI field names
prompt, size, quality, n and background mean what they mean in the OpenAI Images API.
A task, not a held connection
The POST returns an ID right away, so a slow GPT Image 2.5 render at max never holds your connection open.
Images go in as URLs
GPT Image 2.5 references and masks are image_url objects pointing at files you host.
Up to ten per call
Set n and every GPT Image 2.5 image comes back on one task, billed per image.
Usage on the task
Official GPT Image 2.5 tasks report their input and output tokens, so you can check the charge.
Prepaid, no expiry
One balance covers every model on TokLab, and unused credit stays.
What people make with GPT Image 2.5
Product shots
Generate the scene and keep the product. For repeated edits, GPT Image 2.5 Sunburst is the pick.
Interface mockups
Screens and app store art with readable text. Proofread every string.
Catalogue variants
One reference, many colourways: GPT Image 2.5 with n set to 10.
Social posts
Square, portrait and wide crops from one GPT Image 2.5 prompt.
Posters and covers
Headline type set in the image, up to 3840 pixels wide.
Features in your app
Let users run GPT Image 2.5 inside your product, billed to your TokLab balance.
Your first GPT Image 2.5 image in four steps
- 01
Create a key
Make an API key in the console and keep it on your server.
- 02
Add credit
Top up from the minimum. Credit does not expire.
- 03
Send the request
POST a prompt and a GPT Image 2.5 model ID to /v1/images/generations.
- 04
Collect the images
Poll /v1/tasks/{id}. When status is completed, the GPT Image 2.5 URLs are in output.data.
GPT Image 2.5 on TokLab at a glance
What you get on all four GPT Image 2.5 IDs.
- Billing
- Standard: set price per image by quality. Official: input and output tokens.
- Input
- Prompt plus up to 16 image URLs and a mask
- Output
- PNG or JPEG, transparent PNG supported
- Request format
- OpenAI Images API fields, async task
- Failed renders
- Not charged
Side-by-side columns for other image models appear once their pages are live.
GPT Image 2.5 limits to know before you build
Text still needs a check
GPT Image 2.5 usually gets short headlines right. Proofread small print and long strings.
Masks guide, they do not cut
Mask edges are not pixel-exact, so check the boundary on close-ups.
Result links are for delivery
Copy the GPT Image 2.5 files you keep into your own storage.
No streaming
GPT Image 2.5 results arrive when the task ends. Max quality and big canvases take longer.
Why run GPT Image 2.5 on TokLab
One key for every model
GPT Image 2.5 shares one key, one task API and one set of error codes with every image, video and language model.
Live prices
Every GPT Image 2.5 rate on this page comes from the billing ledger when the page loads.
No subscription
Pay per render. Credit never expires.
Failures cost nothing
A failed GPT Image 2.5 task takes nothing from your balance.
Two ways to pay
Standard GPT Image 2.5 IDs have a set price per image. Official IDs bill tokens, so their cost climbs with quality.
Batches bill per image
n = 4 is four images. The prompt is counted once.
auto has no fixed price
With quality on auto the model picks the level, so set it yourself when cost matters.
Other image models
Image models with their own page on TokLab.
GPT Image 2.5 is the first image model with a page here. More image models get theirs as they launch on TokLab.
GPT Image 2.5 guides
GPT Image 2.5 tests and walkthroughs from our own renders.
OpenAI bills GPT Image 2.5 by token. We measured the tokens a 1024×1024 image uses at each quality and priced them on OpenAI and on both TokLab routes.
Flare and Sunburst used the same tokens at every quality level we could compare, so they cost the same. Here is where they differed in speed and output.
GPT Image 2.5 FAQ
What is GPT Image 2.5?+
GPT Image 2.5 is OpenAI's image model from September 8, 2026, released as Flare and Sunburst. Both generate from text and edit your images with the same request. On TokLab each has a Standard ID billed per image and an Official ID billed per token.
Which is better, GPT Image 2.5 Flare or Sunburst?+
Neither GPT Image 2.5 model wins everywhere. Flare is quicker for drafts and volume; Sunburst is slower and keeps more of the original when you edit. Work out the prompt on Flare, then run the final edit on Sunburst.
How do I know whether I used Sunburst or Flare?+
You pick the GPT Image 2.5 model by ID, and nothing switches it for you. The task's model field and your usage history both record the ID that ran.
How much does GPT Image 2.5 cost per image?+
At low quality and 1024×1024, GPT Image 2.5 Flare costs $0.021 on the Standard route and about $0.00448 on the Official route at today's rates. Official prices rise steeply at xhigh and max.
Is GPT Image 2.5 free?+
Not on TokLab. GPT Image 2.5 is paid from prepaid credit, with no free credits. Failed renders are free, and credit never expires.
When did GPT Image 2.5 come out?+
OpenAI released GPT Image 2.5 on September 8, 2026, with Flare and Sunburst as separate API models from the start.
What is the GPT Image 2.5 model ID?+
GPT Image 2.5 has no single ID. Use gpt-image-2.5-flare or gpt-image-2.5-sunburst, or add -official to either for token billing. A plain gpt-image-2.5 is not a model on TokLab, so it is refused.
Can GPT Image 2.5 do image to image?+
Yes. Send up to 16 image URLs and GPT Image 2.5 edits them. Add a PNG mask, the same size as the first image, to limit the change to one area.
What is the GPT Image 2.5 max resolution?+
GPT Image 2.5 goes up to 3840 pixels per side and 8,294,400 pixels in total, so 3840×2160 or 2880×2880. OpenAI calls anything above 2560×1440 experimental.
Does GPT Image 2.5 support transparent backgrounds?+
Yes. Send background: "transparent" with PNG output. GPT Image 2.5 rejects a transparent JPEG before anything is charged.
Can I run GPT Image 2.5 locally or in ComfyUI?+
No. OpenAI serves GPT Image 2.5 as a hosted model and has not released weights. A ComfyUI workflow can still call it through an API like this one.
How do I use GPT Image 2.5 in Codex?+
Run codex mcp add toklab --url https://api.toklab.ai/v1/mcp --bearer-token-env-var TOKLAB_API_KEY, start Codex with TOKLAB_API_KEY set, and ask for a GPT Image 2.5 image. The MCP guide covers Claude Code and Cursor too.
Make your first GPT Image 2.5 image
Try GPT Image 2.5 in the Playground on your own balance, then copy the same request into your code.