ParentFull threadtreprinum·GPT-3.5 has 0.5s average first-token latency and Claude3 Haiku 0.4s.View on HN