What Is Z.ai? Zhipu AI and the GLM Models, Explained
Z.ai is the global brand of Zhipu AI, the Chinese lab behind the open-weight GLM models. Here's what Z.ai is, the GLM lineup, and why it matters.
Z.ai is the global brand and developer platform of Zhipu AI, one of China’s leading AI labs and the team behind the open-weight GLM family of models. If you’ve seen “GLM” topping open-model leaderboards or showing up cheaply on API marketplaces, Z.ai is the company behind it. It has become one of the most important names in the fast-moving world of open-weight AI.
The company behind it
Zhipu AI was founded in 2019 as a commercial spin-out of Tsinghua University’s Knowledge Engineering Group, one of China’s most respected AI research labs. The company rebranded internationally as Z.ai, and in January 2026 it listed publicly on the Hong Kong stock exchange — a notable milestone for a frontier model lab. Z.ai is the outward-facing platform; Zhipu is the lab.
The GLM model family
Z.ai’s flagship product is the GLM (General Language Model) family. What sets it apart from many frontier competitors is its commitment to open weights — the core GLM models are released under the permissive MIT license, meaning anyone can download, run, fine-tune, and deploy them.
The lineup spans several tiers and modalities:
- GLM-5 series — the flagship line: large Mixture-of-Experts models built for advanced reasoning, coding, and agentic work, including the latest GLM 5.2 with its million-token context window.
- GLM-4.5 and GLM-4.5-Air — earlier, widely used models; “Air” is a smaller, more efficient variant.
- Multimodal models — vision-capable GLM models, a voice model (GLM-4-Voice), plus image (CogView) and video (CogVideoX) generation.
This breadth — strong text models plus vision, voice, image, and video — makes Z.ai a full-stack AI platform rather than a single-model shop.
How developers access it
Z.ai offers API access to the GLM family through endpoints that are OpenAI-compatible and Anthropic-compatible. That’s a deliberate, practical choice: developers can point existing code written for those APIs at Z.ai by changing little more than the base URL and model name. Because the weights are open, you can also run GLM models yourself — locally via tools like Ollama or on your own infrastructure — or use them through third-party marketplaces like OpenRouter, often at a fraction of the cost of closed frontier models.
Why Z.ai matters
Z.ai sits at the center of three of the biggest trends in AI:
- Open weights at the frontier. While many top labs keep their best models closed, Z.ai ships competitive models under MIT — a major reason the open-vs-closed gap has narrowed.
- Cost. GLM models are typically far cheaper per token than proprietary alternatives, which reshapes the economics of building with AI.
- A more global model landscape. Z.ai is part of a wave of Chinese labs (alongside the team behind Kimi) producing genuinely frontier-class open models, broadening where the state of the art comes from.
The takeaway
Z.ai is the global face of Zhipu AI and the home of the open-weight GLM models — a full lineup spanning text, vision, voice, and media, released under a permissive license and offered through familiar, drop-in-compatible APIs. For developers who want frontier-adjacent capability with the freedom of open weights and low costs, it’s one of the most consequential platforms to know. Explore it at z.ai, and see the flagship in our GLM 5.2 explainer.
Tagged
Keep reading
Chisato · · 6 min read DeepSeek V4-Flash-0731: Benchmarks, Price, What Changed
DeepSeek's retrained V4-Flash-0731 beats its own flagship on nine agent benchmarks at the same $0.14/$0.28 price, with MIT-licensed weights on Hugging Face.
Chisato · · 6 min read LG K-EXAONE 2.0: Korea's 750B Open AI Model
LG released K-EXAONE 2.0, a 750B-parameter Apache-2.0 open model — Korea's largest, built to rival DeepSeek and Qwen. Specs, benchmarks, and the stakes.
Chisato · · 5 min read DeepSeek V4 Release: Specs, Benchmarks, Peak Pricing
DeepSeek V4 graduates from preview to general availability with two open-weight MoE models, an 80.6% SWE-bench score, and new peak-hour API pricing.