This would be a valuable benchmark if it included languages other than Java, and let me see which models are best at the languages I work with.

My real-world usage does not line up with these results, but I'm not working with Java.