The model overwriting is not working for me during the run creation

Hi All,

I experienced an interesting case.

I want to try the Assistant instance overwriting to be sure it is working as I expected.

Everything is working well, except the model overwriting!
If I change the model to some denied one I get the error message: “The requested model ‘gpt-4-vision-preview’ cannot be used with the Assistants API.”

This is ok, but if I wanna change the model to a GPT4 (gpt-4-1106-preview) one (I am on gpt3 (gpt-3.5-turbo-0125) model originally with the assistant instance) it seems work, but in the result I see the opposite.

So it always answers something like this: I am an AI based on OpenAI’s GPT-3 model. If you have any questions or need help with something, feel free to ask!

Of course, if I test this with an Assistants instance which is created with gpt4 model (gpt-4-1106-preview), it answers it is a gpt 4 model :slight_smile:

I went some rounds with this…

The Response object shows the right one is set…
So in the same response there is it is a gpt4 model, but in the the answer it says no, it is not :slight_smile:

Do you think buggy the model overwriting?
Or is there any hidden rule about this?
Is there any semi-detailed documentation about it?

Thanks for the answers!

I have just tried, if the original model for the assistance instance is GPT4 then it is possible to change the model to gpt3 one. But even after this the GPT3 does not work.

So I hope this is a bug, it seems there is a solution.

You can’t rely on the model telling you what model it is running.

There’s a high chance of what’s known as ‘hallucinations’ or wrong facts…

Thanks, thats good! :slight_smile:

Then I will rely on the model in the response. I hope the invoices too :slight_smile: