A lot of the performance of these open source models might come from distilling the closed frontier models. If those can't raise the funds anymore to train newer and better models then the whole improvement cycle might slow down.