フリー問題

問題文

A team is estimating the cost of a generation-heavy pipeline. Which statement about how tokens are counted is correct?

選択肢

1. Token counting is disabled when the warehouse is suspended, and the rows that ran while it was down are billed at zero on that same statement.
2. The billed token count always equals the token count of the text supplied, and the counting function gives the exact figure that appears on the invoice for that call.
3. For functions that generate new text, both input and output tokens are billable, and the task-specific functions add a prompt to the input so the billed count exceeds the text supplied.
4. Only input tokens are billable for every AI function, because the output is produced by the same forward pass and is therefore not metered separately, whatever the model.