None of the open weight models are really as good as SOTA stuff, whatever their evals says. Depending on the task at hand this might not actually manifest if the task is simple enough, but once you hit the threshold it's really obvious.