# Calculate Streaming Token Usage Costs Implement a usage calculator for language-model API requests. Input and output tokens have different prices, and output usage may arrive through multiple streaming chunks. ```python def calculate_cost( requests: list[dict], input_price_per_million: int, output_price_per_million: int, ) -> tuple[int, int, int]: ... ``` Each request contains `input_tokens` and either `output_tokens` for a non-streaming response or `stream_chunks`, a list whose elements contain the number of newly reported output tokens in that chunk. Return total input tokens, total output tokens, and total cost in millionth-of-a-currency-unit so the computation is exact: ```text cost = input_tokens * input_price_per_million + output_tokens * output_price_per_million ``` ## Constraints and Clarifications - Token counts and prices are nonnegative integers. - A request supplies exactly one output representation. - Streaming chunk counts are deltas, not cumulative totals. - Empty streams and empty request lists are valid. - Do not use binary floating point for billing. - Reject malformed negative counts rather than silently correcting them. ## Hints - Normalize both response forms to one per-request usage record. - Keep aggregation separate from formatting into human-readable currency. - Test zero prices, mixed response forms, many tiny chunks, and very large totals.