GLM-5 vs DeepSeek V4 Pro (xhigh)

Two open models, both under two cents a script, neither anywhere near the top of my board. That makes it a genuinely fair fight. GLM-5 wins it, though I want to be honest about the margin. The average scores are close, 82.42 against 80.50, with enough spread that they nearly touch. But the rubric-derived Elo tells a cleaner story: GLM's confidence interval sits fully above DeepSeek's across the same scored task set. The gap I can point at is visual cues, 83.79 versus 73.61. GLM gives my editor usable [SHOW:] directions; DeepSeek's read like afterthoughts. Both share the same bad habit of leaving raw stats in the spoken lines, GLM just does it less. At this price neither is a final-draft model for me, they are drafting engines. If I am picking a drafting engine at a cent or two per script, GLM is the one I load first.

Pick GLM-5 if you want the stronger budget open-weights draft, especially for visual cues your editor can actually use.
Pick DeepSeek V4 Pro (xhigh) if you are already running it and a slightly cheaper, roughly comparable draft is enough.