Zhang Peng
Last reviewed
Jun 8, 2026
Sources
10 citations
Review status
Source-backed
Revision
v1 · 1,682 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 8, 2026
Sources
10 citations
Review status
Source-backed
Revision
v1 · 1,682 words
Add missing citations, update stale details, or suggest a clearer explanation.
Zhang Peng (Chinese: 张鹏) is a Chinese computer scientist and technology entrepreneur who is the co-founder and chief executive officer of Zhipu AI, one of China's leading developers of large language models and the maker of the GLM and ChatGLM model families. The company, branded internationally as Z.ai and registered as Beijing Zhipu Huazhang Technology, was spun out of Tsinghua University's Knowledge Engineering Group (KEG) in 2019. Under Zhang's leadership it became the first Chinese large language model startup to complete an initial public offering, listing on the Hong Kong Stock Exchange on January 8, 2026.[1][2][3]
Zhang is one of several public figures who share the common Chinese name Zhang Peng; this article concerns the Tsinghua-trained engineer who leads Zhipu AI. He is counted among the founders of the cohort of Chinese model startups often called the "AI tigers."[1][4]
Zhang spent much of his early career inside Tsinghua University's Knowledge Engineering Group, the laboratory led by professor Tang Jie that specializes in knowledge graphs, data mining, and large-scale pre-training. By his own account he worked there for more than two decades, focusing on what he calls "engineering transformation," the practical work of turning academic research into deployable systems. He has summarized the lab's guiding idea as "P2P: Paper to Project."[2][4]
In 2018 he completed a doctorate in the Department of Computer Science and Technology at Tsinghua through the university's Innovative Leading Engineering program, a track aimed at experienced engineers rather than conventional research students. His scholarly work centered on knowledge graphs and large-scale pre-trained models, and he was a principal contributor to several well-known Tsinghua systems, including the GLM series of language models, the academic search engine AMiner, and the XLORE knowledge base. He has authored more than ten papers at venues such as ICML and the International Semantic Web Conference.[5]
For most of Zhang's tenure, Chinese rules discouraged people inside the state research system from starting their own companies. "At that time, it was not allowed for people within the system to start companies," he later recalled. A 2018 policy change opened the door, and after roughly a year and a half of preparation the team registered Zhipu AI in June 2019, which Zhang likened to being "the first to eat the crab," a Chinese expression for a pioneer who takes an untested risk.[2]
The company emerged from Tsinghua's KEG and is credited to a founding group that includes professors Tang Jie and Li Juanzi, who are among its controlling shareholders; Liu Debing serves as chairman, and Tang Jie as chief scientist. Zhang has led the company as chief executive since its founding, overseeing business, research, and operations, and he is routinely described in Chinese and English coverage as a co-founder and CEO. Headquartered at the Tsinghua University Science Park in Beijing, Zhipu adopted the mission "let machines think like humans." It initially built knowledge-graph products, shifted decisively toward large models around 2020, and by the mid-2020s employed well over 800 people, most of them in research and development.[1][3][4]
Zhipu's technology is built on the GLM, or General Language Model, architecture developed at Tsinghua KEG. Introduced in a 2022 paper, GLM uses a pre-training method called autoregressive blank infilling and was designed from the outset to be bilingual in Chinese and English. The lab released GLM-130B, an open bilingual model with 130 billion parameters, in 2022.[4]
The breakthrough in public visibility came in March 2023 with ChatGLM, and especially the open-source ChatGLM-6B, a conversational model small enough to run on consumer hardware. It was downloaded widely and helped establish Zhipu as the most academically rooted of China's model startups. The company followed with later ChatGLM versions and then its GLM-4 flagship in January 2024, expanding into multimodal and voice systems such as GLM-4V and GLM-4-Voice. From 2025 it pushed into agentic and reasoning models, releasing GLM-4.5 in July 2025 and GLM-4.6 in September 2025, many with openly published weights under permissive licenses. In February 2026 it released GLM-5, a model reported at around 744 billion parameters and positioned to compete with frontier Western systems; Zhipu said parts of its training relied on domestically produced Huawei Ascend accelerators, a notable response to US export controls.[1][6][7]
| Model | Released | Notes |
|---|---|---|
| GLM-130B | 2022 | Open bilingual model, 130 billion parameters |
| ChatGLM-6B | March 2023 | Open-source chatbot runnable on consumer GPUs |
| GLM-4 | January 2024 | Flagship general model; later GLM-4-Plus, GLM-4V, GLM-4-Voice |
| GLM-4.5 | July 2025 | Agentic and reasoning flagship, openly licensed |
| GLM-4.6 | September 2025 | Updated agentic model |
| GLM-5 | February 2026 | About 744 billion parameters; trained partly on Huawei chips |
| GLM-5.1 | April 2026 | Incremental update to the GLM-5 line |
Zhang has steered Zhipu toward enterprise and developer customers rather than chasing a consumer hit. About 60 percent of its clients are enterprises, and the company says nine of China's ten largest internet firms use its services; government work, often assumed to dominate, accounts for only around a fifth of its business. He deliberately avoided betting everything on a consumer chatbot the way rival Moonshot AI did with Kimi, declining to "bet everything on a single option" given that Chinese consumers have shown limited willingness to pay.[2]
Zhipu raised capital aggressively as a private company. After early venture rounds it reached unicorn status, and in 2023 it took in roughly 350 million US dollars from Alibaba, Tencent, Ant Group, Meituan, Xiaomi, and HongShan. In May 2024 it raised about 400 million dollars from Prosperity7 Ventures, the venture arm of Saudi Aramco, and during 2025 it drew a series of state-linked and local-government investment funds. In January 2025 the United States added Zhipu to the Commerce Department's Entity List, citing national-security concerns; the company rejected any wrongdoing, and the restriction reinforced its turn toward domestic computing hardware.[1][3]
In April 2025 Zhipu became the first of China's model startups to begin a formal IPO process, initially reported to be targeting Shanghai's STAR Market with a plan to raise on the order of 15 billion yuan. It ultimately listed in Hong Kong, where its listing vehicle, Knowledge Atlas, debuted under the ticker 2513 on January 8, 2026. The offering priced at HK$116.20 per share and raised about HK$4.35 billion, roughly 558 million US dollars, valuing the company near HK$51 billion, about 6.5 billion dollars. The Hong Kong retail tranche was oversubscribed more than a thousand times, and the stock rose on its first day to a market value above HK$57 billion. Zhipu said it would direct about 70 percent of the proceeds to research on its general-purpose models. By some accounts the deal was the first IPO by any foundation-model startup in the world.[1][3][8]
The prospectus underlined how costly the model race has become. Revenue grew from about 57 million yuan in 2022 to roughly 312 million yuan in 2024 and reached about 191 million yuan in the first half of 2025, yet the company posted a net loss of about 2.36 billion yuan over that half-year, with research and development spending of roughly 1.595 billion yuan, much of it for computing power. Commentators noted that Zhipu, like peer MiniMax, was losing well over ten yuan for every yuan of revenue. At its debut the firm also warned of an intensifying global price war in AI services.[8][9][10]
Zhang frames the pursuit of artificial general intelligence as a marathon rather than a sprint, and he has been blunt about prioritizing it over near-term profit. Asked to choose between building a profitable company and one that reaches AGI, he answered: "Of course, the one that achieves AGI. No hesitation." He has said he wants Zhipu remembered as "a pioneer in the history of AGI," and he repeats an internal saying that "no matter how much money we raise or how much money we make, it will be a hindrance on our road to AGI."[2]
He has pushed back on the idea that the listing signals a peak or an exit, describing a public market as "a stage that is much closer to the real market" rather than a place to cash out. On fears of an AI bubble, he argued that "whether it's a bubble has no necessary relationship with whether you go public," and he has observed that Chinese AI investment remains a small fraction, perhaps a tenth or a twentieth, of the level in the United States. As of June 2026 he remains chief executive of Zhipu AI and one of the most prominent figures in China's large-model industry.[1][2]