It may have been how you manipulated your prompt ... word order and format seems to matter a lot in how well these LLMs can answer