Article 1 Prompt Phrasing Shifts Model Performance More Than Tier — and Which LLMs Will Get Your Jokes