DishwasherBench: Can Frontier Models Stack a Dishwasher?
DishwasherBench – Can Frontier Models Stack a Dishwasher?
DishwasherBench is a browser game and benchmark that tests language models on loading a real Bosch dishwasher. Players place items in a 3D physics simulation, then a wash on the Eco 50 cycle is simulated with water jets, droplets, and soil chemistry. Points are awarded for clean items. The benchmark includes seven levels, an MCP server for agents, and verified episode records. Early results show models can play, but scoring for plates and cutlery remains unreliable.
A score today shows whether a player packed well and put cups, glasses and pots in mouth down. It does not yet show whether plates and cutlery were stacked so that they wash clean.