This page gives you a quick introduction to GLM. GLM is the large-model series from Chinese AI company Zhipu — operating externally under the English brand Z.ai since 2025, previously known as “Zhipu AI.” The company was founded in 2019 as a spin-off from Tsinghua University research, and its founding team is largely composed of Tsinghua professors. One distinction worth keeping clear from the start: GLM is the model; Zhipu (Z.ai) is the company.
The most notable thing about Zhipu is that it listed on the Hong Kong Stock Exchange in early 2026 (ticker: 02513.HK), becoming the first of China’s “big-six” large-model companies to successfully complete an IPO. It built its following through the open-source GLM ecosystem, serves primarily enterprise and government customers, and has a product line spanning language models, code, multimodal, and agents.
One line to remember it by: a Tsinghua-lab spin-off that built its name on the open-source GLM ecosystem and became the first Chinese LLM company to IPO — then shifted to domestic compute after the US export-control listing.
A few things to hold in mind: Zhipu is public, but still in heavy-investment, net-loss territory — revenue is growing quickly and so are the losses. The ARR and benchmark numbers it publishes are company framing, distinct from recognized revenue in its financials. On the export-control listing, the US and Zhipu give conflicting accounts.
Key Data Snapshot
The key figures in one place. Zhipu is listed and has public financials, but some of these are official framing or forward-looking figures — we try to label them clearly.
| Item | Data |
|---|---|
| Model / Product name | GLM series LLMs (including open-weight and closed-source API versions); also CodeGeeX, CogVideoX, Zhipu Qingyan, and others |
| Parent company | Zhipu (Z.ai, formerly “Zhipu AI”) |
| Founded | 2019 (spin-off from Tsinghua University) |
| Headquarters | Beijing, China |
| Founders / Core team | Tang Jie, Li Juanzi (Tsinghua professors); CEO Zhang Peng; Chairman Liu Debing |
| Company type | Independent AI company listed on the Hong Kong Stock Exchange (02513.HK); Tsinghua-affiliated |
| IPO and market cap | Listed January 8, 2026; raised approximately US$558M; first-day closing market cap approximately HK$57.89B (~US$7.4B) |
| Flagship products | GLM language models, CodeGeeX (coding), CogView / CogVideoX (multimodal), AutoGLM (agent), Zhipu Qingyan / ChatGLM (consumer-facing) |
| Open-source status | Dual track: open-weight GLM versions available + closed-source API and private deployment |
| Primary customers / use cases | Core focus on enterprise and government (B2B / B2G); company says open-source flagship models have accumulated over 40 million cumulative downloads globally |
| Compute variables | After being added to the US export-control Entity List in early 2025, shifted to domestic compute such as Huawei Ascend |
Two reminders when reading these numbers. ① Zhipu’s ARR (forward annualized) and the revenue actually recognized in its financials are not the same thing — when you see an “ARR of X billion,” first check whether it’s an annualized projection or actual booked revenue. ② Its revenue is growing fast but it remains deeply unprofitable, burning heavily on R&D. Keep “growth trajectory” and “profitability” as separate judgments.
A Quick Tour of Seven Dimensions
Seven dimensions are a useful frame for understanding an AI company. This page gives the overview; Penchan will break the key dimensions into standalone pieces over time.
① Technology and product direction: The GLM (General Language Model) architecture came out of Tsinghua’s research team and started going open-source from 2020, developing into a dual track of “open weights + closed-source API.” Beyond the language model line, the product suite includes coding (CodeGeeX), multimodal (CogView / CogVideoX), and agents (AutoGLM), with a recent push toward longer autonomous-execution capabilities. The benchmark scores the company publishes are its own framing — treat them as “company claims” when reading.
② Customer structure and market positioning: Unlike Doubao, which skews consumer, Zhipu’s core is enterprise and government (B2B / B2G). Its Tsinghua academic pedigree, private-deployment option, and compatibility with domestic chips are its differentiators. The China LLM competitive field includes DeepSeek, Qwen, Wenxin, Kimi, Doubao, and others — presented here in parallel, without picking a winner.
③ Ecosystem and partnership strategy: It has built a sizable developer community through open source, and its investor base includes significant industry capital — Alibaba, Tencent, Meituan, Xiaomi, among others — as well as Middle Eastern sovereign funds. Model weights published on HuggingFace and similar platforms have expanded its reach.
④ Valuation and financial model: As a listed company, Zhipu has public financials. The story is “high growth, still losing money”: revenue is growing sharply year-over-year while R&D investment and net losses are also large. Its ARR (forward annualized) will be noticeably higher than recognized revenue in the financials — they represent different concepts.
⑤ Commercialization risk and regulation: Several pressures are running in parallel. Customer concentration is relatively high (a small number of clients account for a significant share), and China’s LLM market is in a price war, making gross margin a long-term variable. On the China regulatory side, generative AI requires content-safety registration (a standard compliance backdrop for the industry). These are factual descriptions of risk, not a value judgment on the regulatory regime itself.
⑥ Geopolitics and supply chain: This is the section people ask most about. In early 2025, Zhipu was added to the US Commerce Department’s export-control Entity List — the US cited national-security concerns, Zhipu publicly denied them, and the two accounts remain inconsistent. Treat them as parallel narratives. In practice, because its China-market business doesn’t rely on direct US GPU purchases, the concrete impact has landed more on fundraising, partnerships, and overseas expansion. The company has shifted training to Huawei Ascend and other domestic compute (its image-generation models have been confirmed as trained entirely on Ascend).
⑦ Leadership and governance: Zhipu is a textbook “Tsinghua-affiliated” team — founders Tang Jie and Li Juanzi are Tsinghua professors, CEO is Zhang Peng, and Chairman is Liu Debing. As a public company, its equity and voting-rights structure should be read from its public disclosures; the founding team collectively holds a substantial share of voting rights.
Key Milestones
The pivotal moments that brought Zhipu to where it is today:
| Date | Milestone |
|---|---|
| 2019 | Zhipu founded (spin-off from Tsinghua University) |
| 2020 | GLM pre-training architecture begins going open-source |
| Jan 2025 | Added to the US Commerce Department export-control Entity List (Zhipu publicly denied and objected) |
| Jul 2025 | Rebrands externally as Z.ai; releases GLM-4.5 |
| Jan 8, 2026 | Lists on the Hong Kong Stock Exchange (02513.HK); raises approximately US$558M; becomes the first Chinese LLM company to successfully IPO |
| First half 2026 | Releases next-generation GLM series; publishes its first post-IPO financial report (high growth, still in loss) |
Milestones will continue to be updated; figures are based on the latest announcements (table last compiled: May 2026).
Further Reading and Upcoming Standalone Pieces
Penchan will break the key dimensions into standalone pieces, rolling them out over time:
- Zhipu IPO: Of China’s six big LLM companies, why did it cross the finish line first
- After the Entity List: how Zhipu shifted to domestic compute
- ARR is not revenue: reading Zhipu’s first post-IPO financial report
- How it compares to other domestic models: different paths from DeepSeek and Qwen
- Want to learn how to use GLM or wire up the API? Penchan will cover that in separate tutorials (link to follow once live)
- For the full hardware chain: The AI Hardware Supply Chain, End to End