GPT4o - not sticking to maximum words in prompt?

API
1

I build a lot of chatbots so I immediately started testing gpt4o. It works great with the prompts used with gpt-4-x, and the speed is amazing.

EXCEPT it does not seem to stick to the maximum words.
Let’s say that I set max_tokens to 500, and in the prompt I ask it to stick to maximum 150 words (in a number of prompt variations). → This works well for gpt4, but gpt4o frequently runs out of tokens before it has completed its sentences.

This may have something to do with the new tokenization used by gpt4o, or something else. Please let me know if you find successful prompts that gpt4o sticks to

2

I am facing the same issue with this model… Plus JSON mode seems to not work as good as 4 turbo.

3

Interesting. That was next on my list to test. Do you mean it gives invalid json? Or does it output text before/after the json?