Mei—or, the Models Engineering Index—calculates language models' practical performance on engineering tasks.