A1 · Building With Model APIsCost and Latency BudgetsSet a per-request budget and measure where the time actually goes.Ask for four hundred tokens and you have spent ten seconds. Nothing else in the request comes anywhere close to that.