Kimi K2.7 Code vs. GLM-5.2: which open-weight coding model to self-host on vLLM

Moonshot AI released Kimi K2.7 Code on June 12, 2026, and Z.ai (Zhipu AI) released GLM-5.2 on June 13, 2026, both open-weight Mixture-of-Experts models designed for agentic coding workflows. The models support vLLM and SGLang, and the article compares their architectures, benchmark results, and self-hosting costs to help teams decide which to deploy on their own infrastructure.

Member-only story Kimi K2.7 Code vs. GLM-5.2: which open-weight coding model to self-host on vLLM You’ve just finished reading the sixth “open-source model beats GPT-5.5” post this month, and you’re still no closer to an infrastructure decision. Your team needs a coding agent backbone, your legal department won’t sign off on sending proprietary code to a third-party API, and your cloud GPU budget is real money with real accountability. In June 2026, two serious candidates landed days apart: Kimi K2.7 Code from Moonshot AI on June 12 and GLM-5.2 from Z.ai Zhipu AI on June 13. Both are open-weight MoE Mixture-of-Experts models. Both support vLLM and SGLang. Both are explicitly designed for agentic coding workflows. And both come with benchmark numbers that look impressive on a blog post but require careful reading before you commit eight H200s to them. This article is a hardware-honest, numbers-driven comparison. By the end, you’ll understand the architecture of each model, what their benchmark results actually mean, how to configure vLLM for each one, when self-hosting breaks even against the respective APIs, and which model wins for which use case. Core concepts: what makes these models tick Before diving into the comparison, it helps to understand the shared architectural pattern that defines both…