Skip to main content

OpenAI GPT Image 2.5

Text-to-image generation with five quality tiers, two selectable models, native transparent backgrounds and an auto aspect ratio that lets the model choose the shape.
GPT Image 2.5 is a text-to-image API that turns a written prompt into one or more images. It is the successor to GPT Image 2: lower latency at every tier, two extra quality tiers (xhigh and max), transparent backgrounds as a first-class option, and an auto aspect ratio. It keeps what the previous generation is chosen for — close adherence to long, specific prompts and legible text rendered inside the image. To edit an existing image instead of generating from scratch, see GPT Image 2.5 Edit.

​
Key capabilities

  • Five quality tiers: quality accepts low, medium, high, xhigh and max. The tier changes render quality and latency, so the same prompt can be drafted cheaply and finalised at the top of the scale
  • Two models in one endpoint: variant selects flare (latency-optimised, the default) or sunburst (editing-precision). Same parameters, same price
  • Auto aspect ratio: aspect_ratio defaults to auto, which lets the model choose the shape instead of forcing a preset
  • Transparent backgrounds: background: "transparent" is native. Use it with png or webp, the two formats that carry an alpha channel
  • Batch generation: request up to 10 images in a single call with num_images. Each image in the batch is billed
  • Async processing: every request returns a task ID immediately. Poll the task endpoint or supply webhook_url to be notified on completion

​
Use cases

  • Marketing and advertising: on-brand key visuals where the headline or product name has to be readable inside the image
  • Logos, icons and overlays: assets that need a real alpha channel rather than a white box behind them
  • Packaging and print mockups: labels, boxes and signage that need real text rather than text-shaped noise
  • Product and UI concepting: interface screens, dashboards and device mockups with legible captions
  • Content pipelines at scale: batch variants at low, review, then re-render the winner at max

​
Generate images with GPT Image 2.5

Submit a prompt to create a generation task. The API responds with a task ID; collect the result by polling or via webhook.

POST /v1/ai/text-to-image/gpt-image-2-5

Create a new image generation task

GET /v1/ai/text-to-image/gpt-image-2-5

List all GPT Image 2.5 tasks with status

GET /v1/ai/text-to-image/gpt-image-2-5/{task-id}

Get task status and results by ID

​
Parameters

​
Pricing

The quality tier sets the price per image and resolution multiplies it: Every image in num_images is billed, so a 4k request for 4 images costs twelve times a single 1k image of the same quality tier. xhigh and max are currently billed at the high price. The endpoint reference pages above are generated from the OpenAPI specification and are the authoritative, complete parameter list.

​
Frequently Asked Questions

GPT Image 2.5 is a text-to-image API powered by OpenAI’s GPT Image 2.5 model line. You submit a text prompt and receive a task ID immediately. Poll the task endpoint or configure a webhook to receive the generated images when processing completes.
auto is the default and lets the model choose the shape of the output instead of forcing one of the presets. It is most useful on the edit endpoint, where the output then keeps the aspect ratio of the images you sent, including ratios that are not in the preset list. Note that auto only leaves the size free at resolution: "1k"; at 2k and 4k it renders a square at that tier.
They are the two models of the GPT Image 2.5 line, offered on the same endpoint with the same parameters and the same price. flare is optimised for latency and is the default; sunburst is the editing-precision end of the line. Requests that do not set variant always resolve to flare.
Send background: "transparent" together with an output_format that has an alpha channel — png or webp. Asking for a transparent background with jpeg returns a 400, because JPEG cannot carry transparency and silently flattening it would corrupt the asset you asked for.
They extend the scale past high, spending more provider compute for the most demanding renders. They are currently billed at the high price.
GPT Image 2.5 is faster at every tier, adds the xhigh and max tiers, exposes a selectable model through variant, supports transparent backgrounds natively, and defaults to an auto aspect ratio. GPT Image 2 remains available with its three tiers and its square_1_1 default.
Rate limits depend on your subscription tier. See Rate Limits for current limits.
Cost per image depends on the quality tier and the resolution multiplier. See the Pricing page for current rates and subscription options.

​
Best practices

  • Tier the workflow: draft at quality: "low", review, then re-run the winning prompt at a higher tier. The prompt does not need to change
  • Leave the shape to the model: keep the default aspect_ratio: "auto" unless the output has to fit a fixed frame, and pick a preset when it does
  • Batch deliberately: num_images is billed per image. Batch at low for exploration, generate single images at the top tiers
  • Write text explicitly: put the exact string you want rendered in quotes inside the prompt, and keep it short
  • Pick the format before the background: a transparent render needs png or webp
  • Production integration: use webhook_url instead of polling for scalable applications
  • Error handling: implement retry logic with exponential backoff for 503 errors
  • GPT Image 2.5 Edit: edit existing images with the same model line, tiers and variants
  • GPT Image 2: the previous generation, with three quality tiers
  • Nano Banana Pro: Google Gemini image generation, also strong at in-image text
  • Seedream 5 Pro: high-resolution text-to-image with a generate and edit split