Zhipu AI

Zhipu AI (operating internationally as Z.ai) is a Chinese artificial intelligence company headquartered in Beijing, spun out of Tsinghua University in 2019 and best known for developing the GLM family of large language models, including the ChatGLM series and the GLM-4 and GLM-5 generations. The company is regarded as one of China's "AI tigers" alongside Moonshot AI, Baichuan, and MiniMax, and in January 2026 it became the first pure large language model company anywhere to complete an initial public offering, listing on the Hong Kong Stock Exchange as Knowledge Atlas Technology (2513.HK).^[1][2]

History and Background

Zhipu AI was established in 2019 at the Tsinghua University Science Park in Beijing by professors Tang Jie and Li Juanzi, both members of the university's Knowledge Engineering Group (KEG). The company initially focused on knowledge graph technology before shifting to large-scale language model research in 2020, raising its first significant funding round of nearly US$15 million in September 2021.^[2]

The company's research lineage traces to the GLM (General Language Model) pretraining framework released by the Tsinghua group in March 2021, which proposed an autoregressive blank-infilling objective as an alternative to BERT-style masked language modelling and GPT-style causal decoding.^[3] In August 2022 the team released GLM-130B, a bilingual 130-billion-parameter model trained on 400 billion tokens, which became a widely used open alternative to GPT-3 in Chinese research and development communities.^[3]

In 2023 Zhipu launched ChatGLM-6B, a compact open-weight variant released under an Apache 2.0 licence that could run on consumer hardware with roughly 6 GB of VRAM, catalysing a domestic ecosystem of developers and enterprises building on the model.^[3] The GLM-4 generation followed in January 2024, and the company rebranded its international operations as Z.ai with the z.ai domain as it expanded overseas.^[4]

In January 2026 Zhipu listed on the Hong Kong Stock Exchange, raising approximately US$558 million at a valuation of around HK$51.8 billion (about US$6.7 billion), the first IPO by a pure LLM company globally.^[1][5] The listing was followed by the GLM-5 generation — GLM-5 in February 2026, GLM-5.1 in April, and GLM-5.2 in June 2026 — with GLM-5.2, a mixture-of-experts model with 744 billion total parameters and 40 billion active parameters under an MIT licence, topping open-source rankings on agentic coding benchmarks. The stock subsequently surged, briefly lifting the company's market capitalisation above HK$1 trillion (about US$128 billion) in July 2026 before settling lower.^[1][5]

Key Concepts and Technology

GLM architecture: The General Language Model framework combines autoregressive blank infilling with multi-task pretraining, allowing a single unified model to handle both natural language understanding and generation tasks effectively.^[3]

Mixture of Experts (MoE): Later GLM generations, including GLM-5.2, employ sparse MoE designs that activate only a fraction of total parameters per token — 40 billion of 744 billion — reducing inference cost while scaling total capacity.^[1]

Open-weight strategy: Zhipu has released many GLM models under permissive licences including Apache 2.0 and MIT, positioning open weights as a core distribution channel, though not all of its models, datasets, and enterprise systems are open source.^[4]

Model-as-a-Service (MaaS): The company's commercial model centres on selling API access and enterprise solutions to businesses, with CEO Zhang Peng estimating break-even by 2026 or 2027 amid intense competition and rising compute costs.^[2]

GLM Coding Plan: Z.ai markets flat-rate coding subscriptions (Lite, Pro, and Max tiers at US$18, US$72, and US$160 per month) supporting tools such as Claude Code, Cline, and OpenCode, positioning GLM models as a low-cost alternative to Western frontier models for developer workloads.^[6]

Applications and Impact

GLM models are deployed across enterprise sectors in China including governance, manufacturing, energy, finance, internet services, communications, education, and consumer electronics.^[3] Internationally, Z.ai has positioned itself as a low-cost provider of coding and agentic AI capabilities, with GLM-5.2 scoring 62.1 on SWE-bench Pro and leading open-source rankings on agentic benchmarks at its release.^[1] The company also announced plans for a ~1 GW data centre built on Chinese chips as part of its hardware strategy, and stated intentions for a dual listing in Shanghai to fund continued development toward artificial general intelligence.^[1][5]

>See Also

• ChatGLM
• DeepSeek
• Qwen
• Open Weights
• Mixture of Experts
• Chinese AI Regulation

References