I have set Azure OpenAI on my data, Chat with Azure OpenAI models using your own data. My goal was to reduce token usage in each request.

However, I have noticed additional prompt_token usage, even when I am sending user content without any prompt. For example, if I only send the text hello there, it results in a total of 2628 tokens, whereas it should only be 24. If a longer text (7 words) is provided without any prompt, it results in a total of approximately 3.4k tokens.

Example:

[{'role': 'system', 'content': ''}, {'role': 'user', 'content': 'hello there'}]
total_tokens: {'completion_tokens': 24, 'prompt_tokens': 2604, 'total_tokens': 2628}

----------------------------------------------------

[{'role': 'system', 'content': ''},
{'role': 'user', 'content': 'I worked overtime what should I do?'}]
total_tokens: {'completion_tokens': 52, 'prompt_tokens': 3334, 'total_tokens': 3386}

As you can see, under total token I am seeing prompt_token usage close the ~3.5k.
Where does prompt_token usage come from since I do not provide any prompt or system message? Isn’t the whole purpose of using Azure OpenAI models with your own data to reduce token usage? For each request, additional 3.5k token is very expensive. Will it affect my billing, where would prompt_token be considered as part of input tokens?

Pricing/details/cognitive-services/openai-service/, states that Input (Per 1,000 tokens) is $0.0025, than as I understand for 4,000 tokens, the cost should be $0.01.