20 ms extra, for models which respond in 5 minutes
Right It is a use case where humans are not latency sensitive