We trying to deploy a GPT4 model, but seem to get a GPT3 model instead

Question

We are trying to deploy a GPT4 model, but seem to get a GPT3 model instead. Within the chat functionality of the model it says it has a knowledge cut-off of September 2021 even though the deployed model is a gpt-4 turbo model with knowledge cut-off at April 2023. Also, it has a token limit of 8k instead of 120k.

Are the requests routed to an older model even though we can select GPT4 via the Azure portal interface?

Answer 1

@Christian Moreau | statworx I've check with the team and the model is indeed correct. Asking the model itself about the cut-off is not expected to work but you should be able to get details about events post September 2021.

Here is an example

PS: An FAQ for the same has been added in the docs.