How to count tokens when using context caching

Gemini API
1

Hi all,

When using context caching, I noticed promptTokenCount always returns tokens from my prompt + cached content (e.g. a file content). But for billing estimative purpose, can I consider just countTokens() as the input tokens?

Tks.

2

Hi @Alexandre_Makiyama , for billing purpose itā€™s essential to consider ā€˜promptTokenCountā€™ because billing is calculated based on total number of token processed by the model. But cached tokens are billed at a lower rate when included in subsequent prompts.

3

Hi @GUNAND_MAYANGLAMBAM, thanks for the reply.

I still didnā€™t get it. All subsequent prompts always return ā€˜promptTokenCountā€™ as the total number of tokens, including the file content. Isnā€™t the main purpose of context caching to avoid sending the file content (for example) repeatedly?

Edit: When I say ā€œavoid sending the file contentā€, I mean to avoid counting the file content as tokens every prompt.

4

The pricing page might help: Harga Gemini API  |  Google AI for Developers
The tokens count is the same, caching or not. What changes is, for second, third, etc. request, when caching is in effect, the multiplier ($/token) is significantly reduced (for the cached tokens).

Hope that helps explain it.

5

Ok. Got it.

Thanks a lot.