can some one confirm the total tokens utilised when using realtime model for calling purposes?