Z.AI (Zhipu AI)

Z.ai, formerly known as Zhipu AI and operating in China as Knowledge Atlas Technology (HKEX: 2513), is a Chinese artificial intelligence company founded in 2019 as a spinout from Tsinghua University. The company is headquartered in Beijing and develops the GLM family of large language models, the CogView image-generation models, and the CogVideoX video-generation line. Z.ai completed its Hong Kong IPO on January 8, 2026, becoming the first pure-play foundation-model developer to complete a public listing globally and the first of China's "AI tigers" to go public.

At a glance

• Founded: 2019 in Beijing as a spinout from Tsinghua University's Knowledge Engineering Group.
• Status: Public. Listed on the Hong Kong Stock Exchange (HKEX: 2513) as Knowledge Atlas Technology since January 8, 2026.
• Funding: Approximately $1.4 billion raised in private rounds prior to IPO. January 2026 IPO raised approximately $558 million at debut.
• CEO: Tang Jie (co-founder; Tsinghua University professor and Director of the Foundation Model Research Center at Tsinghua's AI Institute).
• Other notable leadership: Li Juanzi (co-founder; Tsinghua professor; co-led the founding research direction).
• Open weights: Yes, partial. ChatGLM, GLM, CogView, and CogVideoX lines have been released open-weights; the closed-weights GLM-4 multimodal flagships and the closed-weights GLM-5 commercial flagship are gated through Z.ai's API and through enterprise channels.
• Flagship models: GLM-5 (February 2026, 745-billion-parameter mixture-of-experts), GLM-4.5 (July 2025, 355-billion-parameter open-weights agentic model), CogVideoX video generation.

Origins

Zhipu AI was founded in 2019 as a commercial spinout from Tsinghua University's Knowledge Engineering Group, led by Tsinghua professors Tang Jie and Li Juanzi. The founding research direction combined knowledge graphs (the academic background of the founders) with the emerging large-language-model paradigm. Tang Jie has been a long-standing figure in Chinese AI academia, including a leadership role at the Beijing Academy of Artificial Intelligence where he contributed to the 2021 Wu Dao 1.75-trillion-parameter model that was among the largest publicly announced research efforts of its era.

The first ChatGLM open-weights releases in March and April 2023, including ChatGLM-6B and the larger ChatGLM-130B, established Zhipu as one of the most-watched Chinese open-weights developers. The releases were quickly adopted by Chinese-language developer communities and became a reference Chinese-language open-weights option through 2023 and 2024.

The GLM-4 family in 2024 expanded the lineup with closed-weights multimodal capability, and the open-weights GLM-9B and 34B variants kept the developer-community presence intact. The CogView image-generation and CogVideoX video-generation lines were released open-weights through 2024 and 2025, building Zhipu's multimodal credentials beyond text. By late 2025, Zhipu was widely characterized as one of the "Six AI Tigers" of Chinese frontier AI alongside Moonshot AI, MiniMax, Baichuan, 01.AI, and StepFun.

The international rebrand to Z.ai in 2025 paired with a scale-up of international developer-facing positioning. The Hong Kong IPO on January 8, 2026, raising approximately $558 million and pricing at HK$116.20 per share, made Z.ai the first pure-play foundation-model developer globally to complete a public listing. The listing was followed by a 173 percent share-price gain over the subsequent month, peaking at HK$317.80 amid retail enthusiasm and analyst upgrades.

GLM-5 launched on February 11, 2026, as the post-IPO flagship release. The 745-billion-parameter mixture-of-experts model with 44 billion active parameters drove a 28.7 percent single-day share-price gain on the release.

Mission and strategy

Z.ai's stated mission is to develop foundation-model artificial intelligence with a focus on agentic applications and on enterprise and government deployment. Tang Jie has articulated the strategic position publicly as a combination of academic research depth, open-weights distribution, and a paid-API and enterprise-services commercial model.

The strategy combines three threads. First, foundation-model research that has produced the GLM family and the multimodal CogView and CogVideoX lines, with deep integration of knowledge-graph and reasoning-research lineage from the founders' academic work. Second, open-weights releases of mid-tier and earlier-generation models combined with closed-weights gating of the commercial flagships, balancing developer-community adoption with paying-customer monetization. Third, enterprise and government deployment through Z.ai's MaaS platform, which reports 2.9 million users and 12,000 enterprise clients including 50 percent of China's top-10 internet firms.

The competitive premise is that the combination of academic research depth (the Tsinghua origin), public-company capital access (the IPO listing), and government-and-enterprise distribution channels produces a structurally distinctive position in the Chinese AI market. The closed-weights GLM-5 and the open-weights GLM-4.5 form a tiered product line that captures both developer-community goodwill and frontier-tier commercial revenue.

Models and products

• GLM-5. Released February 11, 2026. 745-billion-parameter mixture-of-experts architecture with 44 billion active parameters. Closed-weights commercial flagship. Distributed through Z.ai API and enterprise channels.
• GLM-4.5. Released July 2025. 355-billion-parameter open-weights agentic AI model with permissive licensing for commercial use. Built specifically for agentic workflows including tool use, function calling, and long-horizon execution.
• GLM-4.1V-9B-Thinking and GLM-4.5-Flash. Smaller open-weights variants in the 2026 portfolio targeting developer and edge-deployment use cases.
• GLM-9B / 34B / 10B. Earlier open-weights releases in the GLM family across 2023 to 2025.
• GLM-4 multimodal. Closed-weights multimodal models released through 2024.
• CogView image generation. Open-weights image-generation line, with CogView-3 and successor variants.
• CogVideoX video generation. Open-weights video-generation line, with 5-billion and 2-billion-parameter variants released across 2024 and 2025.
• Z.ai MaaS (Model-as-a-Service) platform. Enterprise-facing commercial platform serving 12,000 enterprise clients, including approximately 50 percent of China's top-10 internet firms.

The commercial channels are the Z.ai developer API, the MaaS enterprise platform, and Hugging Face for open-weights distribution. Z.ai also operates a consumer-facing chat product in China.

Benchmarks and standing

Z.ai's GLM-5 release in February 2026 was framed in industry coverage as a credible challenger to DeepSeek and to the Alibaba Qwen line on the Chinese open-and-closed-weights frontier. Standardized benchmark coverage of GLM-5 has been less widely disseminated than DeepSeek and Qwen coverage, partly because of the recency of the release and partly because the closed-weights flagship is not directly evaluable on open leaderboards.

GLM-4.5 has been characterized in industry coverage as a top-tier open-weights agentic model, with reported third-rank position on global open-weights agentic capability benchmarks at release. The model has been adopted in agentic-coding and tool-use workflows by enterprise customers and by community developers.

Z.ai's standing rests on the GLM family's track record of open-weights releases across multiple modalities, the IPO completion as a financial validation of the business model, the 132 percent revenue-growth trajectory reported in pre-IPO disclosures, and the founding-team research credentials at Tsinghua. The enterprise-customer count and the MaaS-platform user base indicate commercial traction that is unusual among the 2024 to 2026 Chinese AI Insurgent cohort.

Leadership

As of April 2026, Z.ai's senior leadership includes:

• Tang Jie, Chief Executive Officer and co-founder. Professor at Tsinghua University's Department of Computer Science and Technology. Director of the Foundation Model Research Center at Tsinghua's AI Institute. IEEE, ACM, and AAAI Fellow. Previously contributed to the Wu Dao foundation-model program at the Beijing Academy of Artificial Intelligence.
• Li Juanzi, co-founder. Tsinghua University professor. Co-led the founding research direction with Tang Jie.

The company has hired aggressively from Tsinghua and from Chinese AI research organizations. The post-IPO disclosure included the broader senior leadership team in some detail, though specific senior-engineering and senior-research profiles have not been broadly covered in international media.

Funding and backers

Z.ai's funding history includes approximately $1.4 billion in private capital across multiple rounds prior to the January 2026 IPO. Reported pre-IPO valuations included a $3 billion mark in November 2024. The pre-IPO investor base reportedly included Alibaba, Tencent, Xiaomi, HongShan (Sequoia China successor), Chinese sovereign-wealth participation, and prominent venture capital firms.

The Hong Kong IPO on January 8, 2026, raised approximately $558 million at a debut share price of HK$116.20. The post-IPO market capitalization, after the share-price appreciation through the subsequent month, reached the high tens of billions of Hong Kong dollars. Z.ai's prospectus indicated that 70 percent of IPO proceeds would be deployed for foundation-model research and development, with the balance for commercial expansion and working capital.

The IPO is significant beyond Z.ai itself as the first foundation-model-developer public listing globally and as a valuation reference point for the broader Chinese AI Insurgent cohort, several of which have been reported to be exploring follow-on listings.

Industry position

Z.ai occupies a structurally distinctive position among Chinese AI labs. The combination of the academic-research origin at Tsinghua, the open-weights model line across text, image, and video modalities, the commercial enterprise traction through MaaS, the IPO-listed public-company status, and the GLM-5 frontier-tier model release produces a profile that no other Chinese AI lab matches. Industry coverage has frequently characterized Z.ai as the most commercially developed of the Chinese AI Tigers and as the leader on enterprise-and-government distribution.

Strategic risks include the post-IPO public-company governance and disclosure requirements that may constrain operational flexibility relative to private peer labs (DeepSeek, Moonshot AI, MiniMax), the intensifying domestic Chinese competition from DeepSeek and Alibaba Qwen on the frontier-capability tier, and the open question of whether GLM-5 closes the capability gap with the leading open-weights frontier from DeepSeek's V4 line. Strategic strengths include the public-company capital access, the diversified open-weights and closed-weights model portfolio, the deep enterprise and government distribution position, and the founding-team academic research credentials.

Competitive landscape

Z.ai competes with several Chinese and international AI labs:

• DeepSeek. Direct Chinese open-weights frontier competitor. DeepSeek's V4 family is the principal capability benchmark target for GLM-5 and successor models.
• Alibaba Qwen. Direct Chinese open-weights competitor with broader portfolio and cloud-distribution channel through Alibaba Cloud.
• Moonshot AI, MiniMax, Tencent Hunyuan, ByteDance Seed, Baidu, StepFun, 01.AI. Peer Chinese AI Insurgents and incumbents. Z.ai's distinguishing features are the IPO listing, the academic-research origin, and the multimodal portfolio breadth.
• OpenAI, Anthropic, Google DeepMind. Closed-weights frontier competitors. GLM-5 is positioned to compete on the frontier-capability tier; GLM-4.5 competes on agentic open-weights capability.
• Meta AI / FAIR. Direct global open-weights competitor through the Llama family.
• Reflection AI. US-based Insurgent positioning to compete with Chinese open-weights frontier developers in 2026.

Outlook

Several open questions affect Z.ai's trajectory in 2026 and 2027:

• The capability profile of GLM-6 and successor models. GLM-5 set a credible benchmark; sustaining release cadence at frontier-tier capability is the central technical question.
• Z.ai's post-IPO commercial growth trajectory, particularly the MaaS enterprise-platform revenue and user-base expansion.
• The pace of international expansion under the Z.ai brand, including the reported government-customer outreach in markets outside China.
• The extent to which Z.ai's IPO listing influences the listing decisions of peer Chinese AI Insurgents (Moonshot AI's reported Hong Kong IPO consideration is a directly relevant signal).
• US export-control developments affecting Z.ai's compute infrastructure and the broader Chinese AI hardware-and-software ecosystem.
• Continued senior-talent recruitment, particularly given the public-company status and the visibility of competitive lateral moves in the broader Chinese AI talent market.

Sources

• CNBC: The first of China's 'AI tigers' goes public as Zhipu climbs in Hong Kong debut. January 2026 IPO coverage.
• Bloomberg: China's Zhipu Unveils New AI Model, Jolting Race With DeepSeek. GLM-5 release coverage.
• SCMP: China's Zhipu AI launches new major model GLM-5 in challenge to its rivals. GLM-5 strategic context.
• FinancialContent: Knowledge Atlas (HKEX: 2513), The GLM Architect and China's AGI Race. Public-company financial deep dive.
• Pandaily: Zhipu AI's Rise: From Tsinghua Lab to China's First Foundation Model IPO. Origins and IPO context.
• Wikipedia: Z.ai. Comprehensive company history reference.