Zhipu AI

Zhipu AI, known internationally as Z.ai since 2025, is a Chinese artificial intelligence company that develops the GLM (General Language Model) family of large language and multimodal models. Founded in 2019 as a commercial spin-out of Tsinghua University's Knowledge Engineering Group, the company is one of China's most prominent independent model developers and became the first major Chinese large-model company to go public, listing on the Hong Kong Stock Exchange in January 2026.^[1][2]

Zhipu is commonly grouped among China's "AI tiger" start-ups alongside Moonshot AI, Baichuan and MiniMax, and its open-weight releases have given it an international developer following that extends well beyond the Chinese market. The company positions its models as a full model-as-a-service stack — language, vision, video and voice models plus agent tooling — sold to enterprises, governments and developers.^[2][3]

History

Zhipu's origins lie in Tsinghua University's Knowledge Engineering Group (KEG), a research laboratory led by professor Tang Jie that produced some of China's most-cited work in knowledge graphs and natural language processing. The laboratory's AMiner academic search platform, and the GLM research framework developed in 2021, formed the technical base on which the company was incorporated in 2019. Tang Jie and fellow Tsinghua professor Li Juanzi are credited as founders, with Zhang Peng, an alumnus of the same laboratory, serving as chief executive.^[2][3]

The company released GLM-130B in 2022, a bilingual open-source model that established its reputation for large-scale open releases. The 2023 launch of ChatGLM-6B, an early conversational model small enough to run on consumer hardware, made Zhipu widely known among developers worldwide and became one of the most widely downloaded Chinese models of its generation.^[3][4]

Zhipu raised successive private rounds backed by Chinese technology investors, state-linked funds and — uniquely for a major Chinese generative-AI company at the time — foreign capital, when Saudi Arabia's Prosperity7 Ventures joined a funding round that valued the company at roughly US$3 billion. By its listing the company had raised more than US$1 billion in private capital and grown to several hundred employees.^[2]

The product line expanded rapidly: GLM-4 arrived in January 2024, followed by GLM-4-Plus and the end-to-end speech model GLM-4-Voice later that year. In early 2025 the United States Department of Commerce placed Zhipu on its Entity List, restricting its access to American technology — a designation the company has contested. In July 2025 Zhipu released GLM-4.5 and GLM-4.5 Air under a free and open-source MIT licence and simultaneously rebranded its international business as Z.ai. GLM-4.6 followed in late 2025 and drew attention for being trained with Chinese domestic chips, including accelerators from Cambricon, as an answer to export restrictions.^[2][3]

Zhipu listed on the Hong Kong Stock Exchange in January 2026 at an implied valuation of about US$6.6 billion — the first listing by a major Chinese large-model developer. The shares rallied dramatically in subsequent months before giving back much of their gain, making the stock one of the most closely watched AI listings in Asia. The company released GLM-5 in February 2026 and continued a rapid release cadence through the GLM-5 series during the year.^[2][3]

Key Technologies

The GLM family spans large language models, vision-language models such as GLM-4.5V, video generation models under the CogVideo name, and speech systems. Zhipu's design philosophy has been to develop model architectures in-house — rather than fine-tuning Western open models — and to couple frontier-scale training with aggressive open-weight publication of smaller and mid-sized variants.^[2][4]

The company's open-weight strategy distinguishes it from some peers: several GLM releases are published on Hugging Face and GitHub under permissive licences, allowing organisations to download, modify and self-host them. This has made GLM models a common choice for developers who need capable models without per-token API costs or foreign data transfers.^[4][5]

Commercially, Zhipu operates a model-as-a-service (MaaS) platform, offering API access to its models with tool-use, long-context and agent capabilities for enterprise applications. The Z.ai platform serves international developers, while the Chinese domestic business focuses on enterprise and government deployments, including smart-city projects.^[2]

Applications and Impact

Zhipu's models are used for enterprise knowledge management, customer service, document processing, coding assistance and public-sector digitalisation in China, where the company competes with DeepSeek, Alibaba's Qwen team and other major laboratories. Its government contracts — including a reported RMB 61.28 million agreement with the city of Hangzhou in 2025 — illustrate the scale of state-driven AI procurement in China.^[2]

Internationally, the open-weight GLM releases have become part of the global open-model ecosystem that also includes Meta's Llama, Alibaba's Qwen and Mistral's models. Zhipu's listing gave public-market investors their first direct exposure to a Chinese frontier-model developer, and its share price became a bellwether for sentiment toward China's AI sector more broadly.^[2][3]

>See Also

• ChatGLM
• DeepSeek
• Qwen
• ILMU (Malaysian LLM)
• National AI Office Malaysia