GPT Image の出力 token がなぜこんなに多いのですか?

{
  "size": "1024x1024",
  "quality": "low",
  "n": 1
}

{
  "size": "3840x2160",
  "quality": "high",
  "n": 1
}

output tokens = width × height × fixed coefficient

total cost
= text input tokens × text input rate
+ reference-image input tokens × image input rate
+ output image tokens × image output rate

{
  "usage": {
    "input_tokens": 1040,
    "input_tokens_details": {
      "text_tokens": 16,
      "image_tokens": 1024
    },
    "output_tokens": 5146,
    "output_tokens_details": {
      "text_tokens": 0,
      "image_tokens": 5146
    },
    "total_tokens": 6186
  }
}