On billions of records my query costs were generally less than a cent. But I partitioned and clustered on the key fields.