Remco Hendriks

This work introduces MetroLLM-Bench, a 955-case benchmark for testing language models as the policy layer of a transit kiosk, and evaluates twenty-six models from six vendors, of which twenty-three are ranked.