I mean it depends on variety of factors such as the cost generated by each model you are using. For example - if you are gpt 4o rather than llama 8b or something then the cost for gpt4o will be much higher than the llama 8b model