Create a cost table that takes into account: - model name regex - provider (optional) - input token cost - output token cost - audio token cost - cache read token cost - cache write token cost - reasoning(?) token cost Create a spreadsheet that can be used to predict a LLM span's cost