The brief
- GLM-5 (744B) released Feb 2026 topped open rankings; GLM-5.2 (753B) extends to 1M context, surpasses GPT-5.5 at 1/6 cost.
- Hong Kong IPO Jan 2026: $558M raised, $7.4B valuation; market cap reached $62B by August 2026.
- Nearly 7M API users globally. Research team published 50+ papers on models, sparse MoE, and multimodal systems.
Technical approach
MIT-licensed models on Hugging Face
Z.ai publishes weights for anyone to run, customize, and deploy without per-token fees. Addresses data privacy and enterprise lock-in.
Dynamic routing of 744B/753B parameters
Only 40B parameters activate per token. Cuts inference compute while maintaining frontier capability at low pricing.
1M-token window with 2.9× efficiency
GLM-5.2's IndexShare sparse attention enables multi-document reasoning and long-horizon tasks without quadratic scaling.

