Managed inference APIs look simple on the pricing page. Pay per token, no infrastructure to manage, done. Then your usage scales, and the bill starts telling a different story. The hidden costs of managed inference APIs rarely show up in the headline rate — they show up in the fine print,