DishwasherBench – Can Frontier Models Stack a Dishwasher?

DishwasherBench is a benchmark tool that tests frontier AI language models' ability to optimally load a dishwasher by simulating a real Bosch washing machine with physics-based gameplay. The system allows both humans to play in a browser and AI models to interact through an API, with scoring based on how clean items come out after a simulated wash cycle. Currently in version 0.1.0, the benchmark is calibrated on real machine testing but has limitations in accurately scoring plates and cutlery placement.