Setup the API to Replicate a Custom GPT Exactly

API
1

I am struggling to achieve the same responses via the API as I generate in my Custom GPT. The instructions, model, temperature and token limit are all the same. Is there a way to setup the API call to make it the exact same as the Custom GPT?

1 Like
2

Here’s why you cannot:

{'error': {'message': "The requested model 'chatgpt-4o-latest' cannot be used with the Assistants API."

  1. You’d need the full system message of ChatGPT, from “You are ChatGPT”, down to placing internal tools (not functions) like ## guardian_tool (which you cannot)
  2. You’d need the full preamble of a GPT, which adds, You are a "GPT", a custom...
  3. You don’t know the sampling parameters temperature and top_p of ChatGPT. They are high enough that ChatGPT can’t generate the same responses of ChatGPT.
  4. The max_completion_tokens does not affect the type of generation; it just terminates the output if over your budget setting.

What you can do is develop your own product that is better than ChatGPT, based on the best model for the application from those that are on the API, and your own customization of the instructions to suit it.

Starting tip: Use gpt-4.1 as the model, and craft new instructions – understanding that they are the system message that guides the whole operation of the AI, not merely modifications.

2 Likes
3

Thank you for the starting tip. Is there a developer you would recommend that I can work with? I understand the product I am building very well, in terms of input and output and the key logic to generate the responses, but I am starting from scratch as a non-developer.

1 Like