Calculating cost for a single Gemini 2 request with audio and text

Gemini API
,
1

I’m trying to calculate the cost for a single multi-modal request for Gemini 2.0 Flash, because I need to charge this amount to my user.

In the documentation here Vertex AI Pricing  |  Generative AI on Vertex AI  |  Google Cloud it states that

1M Input tokens has a cost of $0.15
1M Input audio tokens has a cost of $1.00
1M Output text tokens has a cost of $0.60

In the SDK, I am returned the following usage metadata:

prompt_token_count
candidates_token_count
total_token_count

My understanding is that if the input was only text, then the cost would be

cost = (0.15 / 1000000) * prompt_token_count + (0.6 / 1000000) * candidates_token_count

My question is:

Does the prompt_token_count include both the text input tokens and the audio input tokens? If so, how do I know how many tokens are due to text input vs audio input?

3

Assuming you’re using the python sdk, this way you should be able to get both text and audio tokens:

    for prompt_tokens_detail in response.usage_metadata.prompt_tokens_details:
        print(f"Media: {prompt_tokens_detail.modality.name} token count: {prompt_tokens_detail.token_count}")