Latency

How long one user waits: the time to first token, then the speed of later tokens. It trades against throughput.