I am new to this api. I am now confused why the cost of intructGPT is much higher than GPT3.5 even if I set the model name to be “gpt-3.5-turbo”.
That refers to use of the model gpt-3.5-turbo-instruct, which is accessed through the completions endpoint. It also is other instruction-tuned models such as davinci-003.
Have you not used that model? $0.38 would be many requests at $0.002 per 1000 tokens, although you can use up to nearly $0.01 per call with maximum input or output.
Chat history also counts if you are using chatbot software, or continue sending all the previous output with “submit” in the playground.