A routed token is any input or output token from a request covered by an AI Router routing decision, regardless of who executes the request. This means the same usage measure applies whether AI Router proxies the request or your application calls the selected model directly.
Usage comes from three sources:
Estimating a missing reading is standard metering practice, similar to how a utility estimates usage when it cannot read a meter. Billing is based on measured usage. Where usage data is missing, we estimate conservatively based on your measured usage profile, which is fixed in your contract.
Estimation fills gaps in the usage record; it is not a penalty for a missing report. Reporting actual usage gives you the most accurate record and reduces the need for estimates.
Usage reports support more than accurate billing. Savings and opportunity reports compare routing decisions with the requests your application actually executed, so those reports are only complete when selection usage is reported back for every routing decision.
See the pricing page for plan details or the pricing FAQ for short answers to common usage and billing questions.