I'm looking ahead to the next wave of open-weight models that are as efficient as DSv4 (which is really efficient), and have been heavily distilled on GLM 5.2 (which is trivial, given it is open weight)