Assistants API Token Inconsistencies

API
1

I’m having trouble with the assistant api. This is how I’m initializing it:

updated_assistant = client.beta.assistants.update(
                    assistant_id=assistant_id,
                    model=model,
                    file_ids=[file_id]
                )

I’m updating the model, assistant id, and file id based on user information. The only relevant thing that changes is the model, I switch between gpt-3.5-turbo-1106 and gpt-4-1106-preview. When using GPT 3.5 Turbo, everything works as expected. I get correct answers from the uploaded file, and the total token cost is only ~1000. When I switch to GPT 4 Turbo, however, I get strange results. While it still answers questions correctly, it consistently uses over 4000 tokens to do so, making it ridiculously more expensive to use than GPT 3.5 Turbo. The code is the same for each model, the only thing that changes is the model used.