Study Finds Switching Models Mid-Session Erases Prompt Cache Savings, Costing Up to 11x

A study found switching models mid-session erases prompt-cache savings, costing 5x at 10K tokens and 11x at 60K tokens.