Blind head-to-head: GLM-5.2 vs DeepSeek V4 Pro on 3 real decisions

Top 3 from our 8-model benchmark (Opus 4.8, GLM-5.2, DeepSeek V4 Pro) were within 1.45 points. GPT-5.5 scored in the same band but we already use it through Codex, so it was left out of this comparison. We asked the top three to propose the optimal stack. All three agreed: keep DeepSeek Pro as primary, keep Opus as escalation, drop GLM-5.2. GLM voted to drop itself, citing a 3-5x rate premium over DeepSeek Pro for a 1.3-point edge inside the noise. We ran a blind head-to-head on 3 real decisions from our ow