GLM-5 was trained entirely on Huawei chips
Summary
GLM-5 by Zhipu AI is a 745B parameter Mixture of Experts model trained entirely on Huawei Ascend chips, featuring a 200K context window and 44B active parameters. It emphasizes agentic intelligence, long-context processing, and open-source ambitions with an anticipated MIT license and ultra-low API pricing, while highlighting US hardware independence. The article also compares GLM-5 to GPT-5 and discusses its IPO-backed funding timeline and potential for enterprise deployment.