Comment by the_duke

Most of the Chinese models get into long complicated thinking loops before they accomplish something more complicated.

This is also true for Deepseek 4(.1) .