Zhipu AI — China's Leading GLM Creator

The First Major Chinese AI Company to Go Public

On January 8, 2026, Zhipu AI (rebranded internationally as Z.ai) listed its shares on the Hong Kong Stock Exchange at HK$116.20, raising $558 million in the largest IPO for a major LLM company in history. Within a month, the stock jumped 173%, and JPMorgan issued an “Overweight” recommendation with a target price of HK$400. The story didn’t begin with the IPO — but in a laboratory at Tsinghua University in 2019.

Founding Story: From Academic Research to “Chinese Tiger”

Zhipu AI was founded in 2019 emerging from Tsinghua University’s research environment — one of the world’s top technical universities. The founders were academic experts and researchers who practiced research in large language models, attracted by the opportunity to turn their research into commercial products.

The company chose to develop the GLM (General Language Model) series based on an innovative architecture that integrates discrimination and generation in a single model — an approach that differs from GPT (generates only) and BERT (discriminates only).

In 2023, it raised 2.5 billion yuan (about $350 million) from Chinese tech giants: Alibaba, Tencent, Meituan, Ant Group, Xiaomi, and HongShan. This backing from major players gave it a launch with infrastructure and commercial relationships preparing it for the enterprise market.

In January 2025, the US Department of Commerce placed it on the entity list due to national security concerns — meaning it was officially banned from purchasing advanced Nvidia H100/H200 processors.

Main Products and Models

GLM Series

The backbone of all the company’s products:

• GLM-3 (2022-2023): The model that established Zhipu’s position in the Chinese market.
• GLM-4 (2024): 60% improvement over GLM-3, with support for 128,000 context tokens and 100% retrieval accuracy in needle-in-haystack tests.
• GLM-4.5 and GLM-4.5 Air (July 2025): New generation announced with the international rebranding to Z.ai.
• GLM-4.5V (August 2025): A vision-language model with 106 billion parameters with MoE architecture.
• GLM-4.6 (September 2025): Trained on local Chinese chips (Cambricon Technologies).
• GLM-4.7 (December 2025): Pre-IPO. Claimed to surpass GPT-5 on LiveCodeBench programming benchmark (84.9%), achieving 73.8% on SWE-bench Verified (highest open-source model), and 95.7% on AIME 2025.
• GLM-5 (February 2026): Built on 744 billion parameters, optimized for programming and intelligent agents, capable of 8 hours of continuous independent work.
• GLM-5.1 (April 2026): Latest version — the only open-source model capable of eight hours of continuous work. Surpassed Anthropic Opus 4.6 on SWE-bench Pro benchmark.

Zhipu Qingyan

A free consumer AI chat app with premium advanced features — the consumer interface for GLM models.

BigModel API Platform

The access platform for developers and companies, offering GLM models via APIs with graduated pricing. Most notable feature: GLM-4.7-Flash and GLM-4.5-Flash models completely free (no trial limits).

GLM Coding Plan

A specialized subscription for developers enabling access to GLM models optimized for programming via tools like Cursor, Continue.dev, and Cline. Reached 242,000 paid developers by early 2026.

GLM-5V-Turbo

A multimodal programming model that adds “eyes” to programming agents — sees the screen and understands visual interfaces to execute visual programming tasks.

Achievements and Numbers

Funding and IPO

• 2023: 2.5 billion yuan (~$350 million) with participation from Alibaba, Tencent, and others.
• May 2024: $400 million (with Saudi Prosperity7 Ventures participation) at $3 billion valuation.
• January 2026 (IPO): Raised $558 million in Hong Kong Stock Exchange IPO — largest IPO for an LLM company.
• Post-IPO Valuation: $6.587 billion, up 173% within a month.

Financial Performance

• 2024 Revenue: $45.37 million.
• H1 2025 Revenue Growth: Rose to 190.9 million yuan (~$27 million) with 325% growth.
• Losses: 2.358 billion yuan in H1 2025 (intensive R&D investment).
• 2025 Full Year Losses: 4.72 billion yuan — reflects intensified model releases.

Market Share and Technical Performance

• Third largest share in China’s LLM market per IDC (~18%), after Baidu and Alibaba.
• Over 300 patents filed in China.
• GLM is one of the most downloaded Chinese AI models on Hugging Face.
• Samsung Galaxy S25 Chinese version includes a specialized GLM model for devices.

Competition and Challenges

Main Competitors

• DeepSeek: Pressures Zhipu with very aggressive API pricing in Q4 2025.
• Alibaba Qwen 3: The strongest commercially with integrated distribution in Alibaba Cloud infrastructure.
• Baidu Ernie: Most experienced in the Chinese enterprise space.

Competitive Advantage

Government Institution Relationships: Zhipu benefits from China’s policy that 70% of government AI spending must go to “first-batch local models” — and Zhipu is one of the first companies approved on this list.

Huawei Ascend and Others: Zhipu is one of the few that developed complete SOTA models on local Chinese chips, strengthening its strategic position in cases of escalating US restrictions.

Challenges:

• US entity list restricts access to best Nvidia processors, hindering training capabilities.
• Massive losses (4.72 billion yuan in 2025) amid accelerating competition from DeepSeek and Ali Qwen.
• Chinese enterprise market expects low prices with high performance expectations — continuous margin pressure.

Future Vision 2026–2027

GLM-5.1 and Pricing: Zhipu raised prices 8-17% with GLM-5.1 launch in April 2026, a significant signal that it is shifting from a “user acquisition via free” methodology to “revenue generation.” Prices remain very low compared to Anthropic ($1.40 vs. $5 per million input tokens).

Expansion in ASEAN and Global South: Zhipu leads an “Independent LLM Alliance” with ASEAN countries and Digital Silk Road countries — a strategy media-dubbed “the GLM Digital Road.”

Long-Term Agent Model: GLM-5 working eight continuous hours, and more autonomous GLM-5.1 — indicate a strategy to excel in “extended agency” long-term tasks, a market no one has captured yet.

Analytical Conclusion

Zhipu / Z.ai carries three strategic axes that give it flexibility most competitors lack: academic (Tsinghua ties), commercial (government and private sector institutions), and national (Chinese chips, Chinese standards, compliance).

The Hong Kong IPO gave it credibility, liquidity, and additional runway exceeding 36 months. Losses are large but acceptable in the race context. The core question is whether the price increase with GLM-5.1 will turn into a clear path toward profitability before reserves are depleted — or will GLM-5.1’s success in independent long programming give it enough competitive advantage to justify a premium in the global market.