Zhipu AI (Z.ai): the Beijing lab behind the GLM models

A Tsinghua University spin-out

Zhipu AI grew out of the Knowledge Engineering Group (KEG) in Tsinghua University’s computer science department in Beijing. The company was founded in 2019 and is headquartered in Haidian, the capital’s main tech district. It describes itself as an independent large-model developer active in research, paid API access and consumer apps alike.

Outside China it goes by Z.ai, so expect to see two names: Zhipu in news coverage and in China, Z.ai on the sites aimed at everyone else.

The GLM family and its offshoots

The core product is GLM (General Language Model), a family of large language models refreshed at a steady pace. Recent versions are pitched at coding and AI agent work: breaking a problem down, running steps, and course-correcting over long tasks.

Around that core, Zhipu also publishes:

• vision and image-understanding models (GLM-V, CogVLM);
• image generation (GLM-Image, CogView) and video generation (CogVideoX);
• specialized models for OCR, speech recognition and text-to-speech.

A deliberate open-weights strategy

What sets Zhipu apart from many rivals is the license. Its flagship GLM models ship as open weights under the MIT license: you can download, modify and use them commercially, even inside a closed product. In practice, that lets a company host GLM in-house without sending data to an outside service. The models live on Hugging Face under the zai-org organization.

One caveat: not every variant is necessarily released. Versions tuned for speed or price appear to be available only through the API, so the individual model card is the reference.

How to access GLM

There are three ways in:

1. The chat assistant Z.ai, to try GLM with nothing to install.
2. The API, documented at docs.z.ai (bigmodel.cn in China), to build GLM into an app.
3. Downloading the weights for self-hosting with vLLM, SGLang or Transformers.

To see how GLM stacks up against other open models, check our open-source models page or the model comparison tool. If you are weighing several Chinese labs, the Moonshot AI profile covers a similar approach.