I am trying to track LLM token usage in my application. I use usage_metadata to do this. I am trying Gemini models. I am seeing the snippet below and I cannot make sense of it. Neither the input or output add up to the total tokens.
"usage_metadata": {
"input_tokens": 7235,
"output_tokens": 44,
"total_tokens": 7368,
"input_token_details": {
"cache_read": 0
},
"output_token_details": {
"reasoning": 267
}
}