ZhipuAI 智谱 (Z.ai)
Company profile
Section titled “Company profile”- Founded: 2019 (incubated from Tsinghua University’s KEG Lab (Knowledge Engineering Group) in the Department of Computer Science; core team includes Professor Tang Jie 唐杰)
- Headquarters: Beijing
- Listed: Hong Kong main-board IPO in 2026 (ticker 2513.HK), the first Chinese LLM company to go public
- Main models:
- GLM-4.5 / GLM-4.5-Air (open-sourced 2025-07-28, MIT License) — #3 globally and #1 in China across 12 mainstream benchmarks
- GLM-5 (released 2026-02) — fully compatible with Huawei Ascend 昇腾 and other domestic chips
- CogVideo / CogView video and image generation
- Agent products: AutoGLM (browser / mobile automation) + GLM-PC (desktop-operation agent) — the “GLM-OS” vision
- Business model: open-platform API (bigmodel.cn) + enterprise customisation + Z.ai consumer products
Strategic positioning: first Chinese AI-native startup IPO + deep national-team ties
Section titled “Strategic positioning: first Chinese AI-native startup IPO + deep national-team ties”Major Chinese AI-native startups include: ZhipuAI 智谱, Moonshot 月之暗面 (Kimi), MiniMax, Baichuan 百川智能, StepFun 阶跃星辰, 01.AI 零一万物 (founded by Lee Kai-Fu 李开复), and DeepSeek (on an independent path).
Zhipu is the first Chinese AI-native startup to list, marking the entry of Chinese LLM startups into the capital-markets phase.
National-team characteristics
Section titled “National-team characteristics”- Tsinghua heritage: Professor Tang Jie’s KEG Lab is a leading Chinese NLP / knowledge-graph research group
- “New-Generation AI Open Innovation Platform” (cognitive-intelligence direction) — anchored at Zhipu
- Deep participation in the xinchuang 信创 ecosystem: GLM-5 compatibility with Huawei Ascend and other domestic AI chips is a direct response to U.S. export controls
- Government-procurement preference: multiple provincial and municipal government-AI and education-AI deployments
Impact of U.S. export controls
Section titled “Impact of U.S. export controls”The U.S. Commerce Department’s BIS added multiple Chinese AI companies to watch / entity lists in 2025 (whether Zhipu is included should be checked against the latest Federal Register). Zhipu’s explicit Ascend compatibility in GLM-5 is a direct response to this pressure: the ability to train and deploy flagship models without NVIDIA chips.
Deep dive: a triple identity — academic heritage, national team, and listed company
Section titled “Deep dive: a triple identity — academic heritage, national team, and listed company”Unique governance structure
Section titled “Unique governance structure”Compared with other Chinese AI-native startups, Zhipu is distinctive:
| Dimension | Zhipu | Moonshot | MiniMax | Baichuan |
|---|---|---|---|---|
| Founder background | Tsinghua academic | Tsinghua + commercial | Academic + commercial | Sogou + Wang Xiaochuan 王小川 |
| Listing status | Listed (HKEX) | Hong Kong IPO prep | Not listed | Not listed |
| National-team depth | Deepest | Medium | Medium | Medium |
| Open-source intensity | Strong (MIT) | Strong (Kimi K2.5 open-sourced) | Medium | Medium |
| Domestic chips / xinchuang | Explicit compatibility | Not explicit | Not explicit | Not explicit |
Being listed implies:
- Public-disclosure obligations → prospectus and annual reports disclose AI governance information
- 2025 annual report due 2026-03-31 → the first annual report with full-year AI business data and governance narrative
- Shareholder pressure → tension between valuation-driven growth and safety investment
Self-regulation posture
Section titled “Self-regulation posture”Similar to other Chinese AI companies, Zhipu has no standalone “safety framework” document:
- Compliance obligations discharged via CAC algorithm filings (multiple GLM-series filings)
- Participation in TC260 standard-drafting (Tang Jie and others are drafting experts on multiple AI standards)
- Open-source model cards (e.g. GLM-4.5) include some safety-evaluation disclosure
- GPAI CoP not signed
But as a listed company, the public-disclosure obligations impose additional governance-transparency requirements on Zhipu — potentially a watershed for governance transparency among Chinese AI companies.
Policy document snapshot
Section titled “Policy document snapshot”| Type | Document | Link | Subpage |
|---|---|---|---|
| Usage policy | Z.ai service agreement + privacy policy | zhipuai.cn | — |
| Technical reports | GLM-4.5 / GLM-5 Technical Report | ModelScope + arXiv | — |
| Open-source models | GLM-4.5 / GLM-4.5-Air GitHub | github.com/THUDM | — |
| Transparency | CAC algorithm filings + HKEX disclosures (2026-03-31 annual report) | cac.gov.cn + hkexnews.hk | — |
Regulatory-compliance posture
Section titled “Regulatory-compliance posture”- China:
- CAC algorithm filings (multiple GLM-series filings)
- TC260-003-2024 compliance
- Deep xinchuang ecosystem participation (GLM-5 Ascend compatibility)
- HKEX listed-company disclosure obligations (the first AI annual report in 2026 will be a key observation)
- United States:
- BIS watch / entity list (current status to be checked)
- Consumer products not offered in the U.S.
- GLM open weights can be downloaded in the U.S.
- European Union:
- Open-source model downloads trigger AI Act Art. 53 obligations
- GPAI CoP not signed
- No standalone action by Italy’s Garante or other DPAs against Zhipu so far
Company posture, in brief
Section titled “Company posture, in brief”- Balance of academic culture and commercialisation: the research DNA of the Tang Jie team remains strong
- Strong national-team identification: xinchuang ecosystem, domestic-chip compatibility
- Capital-markets orientation: post-IPO valuation and earnings pressure will intensify
- Open-source friendly: MIT licensing is among the more permissive choices among leading Chinese companies