Separate input and output
Estimate requests per day, average input Tokens and average output Tokens. Include system prompts, conversation history and retrieved documents.
Estimate baseline demand
Monthly input Tokens = daily requests × average input Tokens × operating days. Estimate output separately. This is a planning estimate, not a bill.
Account for peaks
Record peak concurrency, rate limits and latency targets alongside monthly volume. Repeated requests, retries and caching can change the estimate.
Confirm billing units
Confirm rates per applicable Token unit and separate treatment for cached input, images, audio and video where relevant.