GLM Is the New Hotness, So Let's Test It On the Homelab

Short summary

Testing three GLM model variants from Z.ai on a high-end consumer homelab (RTX 5090) to evaluate local inference viability. Compares the massive GLM-5.2 (753B), the practical GLM-4.7-Flash (30B MoE), and the small GLM-4-9B baseline across real metrics: tool-calling reliability, inference speed, and performance on Coder Agents coding tasks. Core question: which GLM can actually function as an agentic coding assistant without overwhelming single-GPU hardware?

• •Testing three GLM sizes on consumer homelab (RTX 5090): 753B flagship, 30B lightweight, 9B baseline
• •Focus on practical viability—tool-calling correctness, inference speed, and Coder Agents task completion
• •Clear methodology: honest about hardware constraints and what 'runs' actually means for local LLMs

Generated with AI, which can make mistakes.