Assistant API + gpt4o + filesearch uses more tokens then gpt3.5

API
1

Hello, I created an assistant that uses filesearch. I have a system prompt and an user query.

I noticed when using GPT4o for the assistant the tokens used are around 16.000, and when using GPT3.5turbo-0125 with the same prompt uses around 6.000 tokens?

Why the change in tokens? output can be different I know but it is never 10.000 tokens in difference.

Am I overlooking something with GPT4o, assistant API and/or FileSearch?

2

gpt-4 models get 20 chunks back by default from search
gpt-3.5, 5 chunks