And I think it's very likely that with improved methods you could get opus 4.6 level performance on a wrist watch in few years.

You needed supercomputer to win in chess until you didn't.

Currently local models performance in natural language is much better than any algorithm running on a super computer cluster just few years ago.