GPT-5.6 Sol (ultra) vs GLM-5.3

GPT-5.6 Sol (ultra) wins this one, and the intervals say it plainly: #12 on the board at 2177.4 Elo against GLM-5.3 at #24 with 2041.5, a 135.9-point gap, and the confidence intervals don't touch, so the ranking is real. Overall it's 87.97 to 86.83, a 1.14-point gap, but the consistency number matters more here: GPT-5.6's standard deviation is 1.45 against GLM's 6.6, so GLM swings between strong scripts and weak ones while GPT-5.6 barely moves. Credit where due, GLM-5.3 takes two metrics I care about: tone and voice at 87.89 vs 86.8, and hook strength at 89.54 vs 87.69, so its best openings genuinely land. Everything else goes to GPT-5.6, and some gaps are wide: visual cues at 88.55 vs 81.43, length adherence 91.9 to 87.5, substance 89.01 to 85.93. Now the odd part on cost. GLM is open weights with much cheaper tokens, $1.4 in and $4.4 out per million against $5 and $30, yet per script it lands at about eleven cents against GPT-5.6's fourteen, and it's slower getting there. For roughly three cents more per script, you get the steadier, more complete writer.

Pick GPT-5.6 Sol (ultra) if you want the consistent script every time, with clearly better visual cues, length adherence, and substance, at about fourteen cents each.
Pick GLM-5.3 if you want open weights you can run yourself, stronger hooks and tone on its good days, and can accept scripts that swing more from draft to draft.