Distillation has always been a pretty dumb cope since if it were just a matter of data, Gemini or Grok would at minimum be just as good as open models because you can distill open models to your hearts content, and yet even with half a million GPUs Elon's Collosus couldn't even put Grok on the chart.
Anthropic like to talk about distillation because Chinese labs publish the majority of public research, research that Anthropic implement into their own models, so they're forced into trying to differentiate themselves with data instead of tech.
And the thing is in this context even discussing Mythos at all is a false dichotomy, because the actual model you need to compare Mythos to are also classified Chinese model that only Beijing and Chinese institutions can use. The only model that you should compare GLM5.2 is Opus 4.8.
