Every few months, the bar for what counts as a "practical" coding model moves. GLM-5.2 just moved it again. The shift is less about raw benchmark scores and more about what the model actually lets you...