Zhipu AI, which rebranded internationally as Z.ai in 2025, is one of China’s most prominent AI labs. It grew out of the Knowledge Engineering Group at Tsinghua University in Beijing and is associated with that university’s academic AI research, which is why it is often described as a Tsinghua spinout. The company develops the GLM (General Language Model) family and was an early mover in China’s open-model movement, having released the original ChatGLM open-weights models.
The GLM line spans several capability areas. The GLM-4.5 repository, maintained by the zai-org GitHub organization, describes the model as a foundation model for intelligent agents with “355 billion total parameters with 32 billion active parameters,” built as a hybrid reasoning system that can switch between a thinking mode for complex tasks and a non-thinking mode for immediate responses. A smaller GLM-4.5-Air variant is offered alongside it. The series emphasizes three core capabilities: agentic behavior, reasoning, and coding. Through its developer platform the company also offers multimodal and image-generation models.
A defining feature of Zhipu’s strategy is open release. The GLM-4.5 models are published under the permissive MIT license and, in the company’s words, “can be used commercially and for secondary development.” This places Zhipu alongside other Chinese labs such as the Alibaba Qwen team and DeepSeek that compete by giving away capable model weights, in contrast to the closed, API-only approach taken by several leading US labs.
Why business readers should care: Zhipu is one of a small group of Chinese labs producing frontier-class open models, and its permissively licensed GLM family is a serious option for organizations that want capable agentic and coding models they can download, inspect, and self-host rather than rent through a single vendor’s API.