Why are the gpt-4o and gpt-4o-mini models different in total tokens when the prompt is the same?

[sharp gorge]

with the same prompt why the total tokens are different, can anyone explain it?

gpt-4o

  "usage": {
    "prompt_tokens": 928,
    "completion_tokens": 136,
    "total_tokens": 1064
  },

gpt-4o-mini

  "usage": {
    "prompt_tokens": 17208,
    "completion_tokens": 136,
    "total_tokens": 17344
  },

[static oxide]

Did you use vision? The vision model costs the same but token counts differ

[sharp gorge]

yes I use vision in both models, but why only in the gpt-4o-mini model the total tokens are too large, even though the prompt is the same.