Zhipu AI Releases GLM-5 on Hugging Face Official Model Registry

Zhipu AI updated its official model registry on Hugging Face on August 11, 2026, publishing GLM-5, the latest generation of its open-source GLM model family. The registry lists GLM-5 as a text-generation model built for the Transformers library, meaning developers can load it with mainstream open-source tooling.

The model card shows safetensors weights and tags that point to a mixture-of-experts (MoE) architecture, conversational use, and bilingual English and Chinese support. The entry also references an accompanying arXiv paper (2602.15763), indicating that technical details are being published alongside the weights.

The model has already accumulated 155,036 downloads and 2,119 likes on Hugging Face since the update, a strong early signal of community interest in Zhipu's next-generation open model.

That traction matters because GLM-5 arrives as open-weight models close the gap with closed systems on reasoning, coding, and other core tasks. A new flagship release from one of China's leading AI labs adds another serious option for developers worldwide.

The MoE design aligns GLM-5 with the dominant architecture trend among frontier open models, while bilingual support lowers the barrier for both English- and Chinese-speaking developers to deploy it directly.

For the broader ecosystem, GLM-5 is a major open-source supply update. Sustained iteration from Chinese labs has given the global community more choice, and GLM-5 extends that momentum with a modern, readily usable bilingual model.

What to watch next: independent benchmark results, whether Zhipu pairs the open release with commercial API access, and how quickly a fine-tuning and deployment ecosystem forms around the model. With the model card and paper now public, community evaluation should soon reveal GLM-5's real capabilities.