Large language models like Mixtral grade written responses faster, but they aren’t perfect and often grade differently than a teacher would.