Model quality index methodology is as per this comment (can add perplexity using the dropdown): https://news.ycombinator.com/item?id=39014985#39017632

It's a combination of different quality metrics which have Perplexity, overall, not performing as well. That being said, I think we are in the very early stages of model quality scoring/ranking - and (for closed sourced models) we are seeing frequent changes. Will be interesting to see how measures evolve / model ranks change