I think it's activated by default (it's on openai side, not vapi). The current vapi cost estimates for realtime API are very rough and conservative it seems