GPT Image API: OpenAI image models at half the token price.
Generate pictures from a text prompt with the OpenAI Images API format and the same prepaid key you use for text models. You pay for the measured prompt and image tokens of each image: $15 per 1M, half of OpenAI's $30.
Live prices and cost per image
Per-image figures are estimates for a square 1024×1024 picture; larger sizes and higher quality use more tokens, and each request shows its exact charge in your dashboard.
| Model | Price / 1M image tokens | 1024², low | 1024², medium | 1024², high | Status |
|---|---|---|---|---|---|
| gpt-image-2.5-sunburst | $15.00 | ≈ $0.007 | ≈ $0.027 | ≈ $0.105 | LIVE |
| gpt-image-2.5-flare | $15.00 | ≈ $0.007 | ≈ $0.027 | ≈ $0.105 | LIVE |
| gpt-image-2 | $15.00 | ≈ $0.007 | ≈ $0.027 | ≈ $0.105 | LIVE |
| gpt-image-1.5 | $15.00 | ≈ $0.007 | ≈ $0.027 | ≈ $0.105 | LIVE |
Which image model to use
- GPT Image 2 is the general choice: sharp text inside images and flexible sizes up to 4K.
- GPT Image 2.5 Sunburst suits product shots, promo graphics and design assets.
- GPT Image 2.5 Flare suits illustrations, posters and creative artwork.
- quality="low" is the cheapest way to draft and iterate; switch to medium or high for the final picture.
Generate an image in 30 seconds
Create a key, add $5 or more and call POST https://api.model.sale/v1/images/generations. The image comes back as base64 in data[0].b64_json.
curl https://api.model.sale/v1/images/generations \\
-H "Authorization: Bearer $MODEL_SALE_KEY" \\
-H "Content-Type: application/json" \\
-d '{"model":"gpt-image-2","prompt":"A red apple on a white table","size":"1024x1024","quality":"medium"}'import base64
from openai import OpenAI
client = OpenAI(base_url="https://api.model.sale/v1", api_key="YOUR_KEY")
image = client.images.generate(model="gpt-image-2", prompt="A red apple on a white table", size="1024x1024", quality="medium")
open("apple.png", "wb").write(base64.b64decode(image.data[0].b64_json))Common questions
How is an image billed? By the tokens OpenAI's image model reports for that image (prompt plus image output), at one flat rate. A failed generation is not charged.
Is image editing supported? Generation from text (/v1/images/generations) is available now; editing an uploaded image (/v1/images/edits) is not yet.
How long does one image take? Usually 10-30 seconds for low and medium quality; high quality and large sizes can take a minute or more.