GPT-4o cannot properly call custom functions more than half the time

I guess we are stuck using gpt-4-turbo…