GPT Image API pricing: every version on one page

cURL

# GPT Image 2.6 Flare, async: returns a taskId to poll
curl https://api.apimodels.app/v1/images/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2.6-flare",
    "prompt": "A ceramic mug on a wooden table, soft window light",
    "aspect_ratio": "16:9",
    "resolution": "1K",
    "quality": "medium"
  }'
# poll GET https://api.apimodels.app/v1/images/generations?task_id=... until state is completed

Python

from openai import OpenAI

client = OpenAI(base_url="https://api.apimodels.app/v1", api_key="YOUR_API_KEY")

# gpt-image-2-all: synchronous, the official images.generate() contract
img = client.images.generate(
    model="gpt-image-2-all",
    prompt="Product shot of a white sneaker on a beige background, studio light",
    size="1024x1024",
    quality="low",        # 1K low = $0.01; medium $0.025, high $0.07
)
print(img.data[0].b64_json[:80])

What is the most affordable GPT Image API?

On apimodels.app the lowest fixed price is $0.01 per image: gpt-image-2-all at 1K low, and GPT Image 2.5 Flare or Sunburst at 1K low. At 2K, GPT Image 2.5 starts at $0.02 and gpt-image-2-all at $0.025. GPT Image 2.6 bills per token at 10% below OpenAI's official rates; our 1K low test request comes to about $0.0069 at current rates, and its SVG output is a flat $0.0315. Failed requests are never charged, so refusals and errors do not add to the bill.

Does the GPT Image API on apimodels.app work with the OpenAI SDK?

Yes. Set base_url to https://api.apimodels.app/v1 and use your apimodels.app key. gpt-image-2-all is the drop-in choice: images.generate() and images.edit() return synchronously, exactly as the OpenAI SDK expects. GPT Image 2.5 and 2.6 also answer synchronously when you pass an OpenAI-style WxH size; without one they return a taskId to poll.

Which GPT Image version supports mask editing?

Only GPT Image 2.6. It accepts a mask (white marks the area to change, black the area to keep) or a markup image drawn over the first reference, but not both in one request. GPT Image 2, 2 All and 2.5 return a 400 for a mask; on those, name the region you want changed in the prompt.

Am I charged when a GPT Image request fails or is refused?

No. Credits are reserved when the request is accepted and released automatically if it fails, whether the cause is a timeout, an upstream error or a content-policy refusal. Refusals are logged as CONTENT_MODERATION in your call records. In the last seven days they made up 83% to 99% of failed GPT Image 2 and 2.5 requests.

How long does a GPT Image API call take?

Production medians for the last seven days at 1K: about 35 s on gpt-image-2.5-flare, 45 s on gpt-image-2.5-sunburst, and 61 to 62 s on gpt-image-2 and gpt-image-2-all, with p90 between 65 s and 129 s. GPT Image 2.6 returned a 1K low image in about 20 s in our tests, and 2K or 4K in 13 to 25 s. Set client timeouts with headroom above the p90.

Do I need OpenAI organization verification to use GPT Image?

No. You sign up at apimodels.app, create one API key and call any of the six GPT Image models with it; no OpenAI account or organization verification is involved. The same key also works for the rest of the catalogue, from Nano Banana and Seedream to video and language models.