Token Minutes: The Metric We Need for the Fast-Model Era
Speed is becoming a first-class capability in frontier AI. Here is why I think we need a new metric — token minutes — so we can evaluate model work quality without being fooled by wall-clock latency alone.