Hacker News Reader

How do you qualify what makes a model "Mythos class", and how do you reliably test for it?

Presumably a deepswe benchmark, which IIRC puts GLM 5.2 between opus 4.8 and fable.