Great question — answering as the UphillTech team:
This is pretty classic for all LLMs. They can't count words. Some do a better job than others at staying under limits but none of them can count. The recent reasoning models, i.e. o1 and o3 from OpenAI, seem to do a better job. https://community.openai.com/t/gpt-cannot-count-words-why/996739
Full answer by Gabe Spradlin: https://stackoverflow.stackexchange.com/a/79456125 (CC BY-SA 4.0).
Need it done for your business? UphillTech.
