This just depends on the size of your prompt. Seems like you have ~10,000 tokens in your prompt. It "tripled" because that is the accurate cost OpenAI charges; Vapi just passes on the cost.