A test that evaluates how well language models perform as autonomous agents in interactive environments.