CtrlK
Tsinghua University (清华大学, Qīnghuá Dàxué), often abbreviated as THU, is a public research university located in the Haidian District of Beijing, China. Founded in 1911 on the grounds of a former Qing dynasty royal garden, Tsinghua has grown into one of the most prestigious institutions of higher learning in Asia and the dominant academic center for artificial intelligence research and entrepreneurship in China. By the mid 2020s, the university had become so deeply embedded in the country's AI economy that conservative estimates suggested at least 60 percent of founders behind China's top one hundred large language model companies had a Tsinghua tie, whether as graduates, professors, or alumni investors.
Tsinghua's role in the global AI landscape is shaped by several interlocking pieces. Its computer science department is consistently ranked first in China and among the top in the world. Its Yao Class, founded by Turing Award winner Andrew Yao, has produced an outsized share of the senior researchers behind Chinese frontier models. Its faculty cofounded Zhipu AI, one of the country's earliest and most influential foundation model companies. Its alumni founded or lead Moonshot AI, Baichuan Intelligence, MiniMax, Beijing Galaxy General Robot, and dozens of other firms now considered central to the China AI industry. Tsinghua also operates several large dedicated AI institutes, runs the open source GLM and ChatGLM model families, and serves as a key node in policy and governance work through the Schwarzman Scholars program and the Institute for AI International Governance.
Tsinghua College (清华学堂) was established on April 29, 1911 on the site of Qing Hua Yuan, a former imperial garden in northwest Beijing. The school was founded as a preparatory institution for Chinese students who would receive scholarships, funded by the United States' return of a portion of the Boxer Indemnity, to study in American universities. This founding circumstance gave Tsinghua an unusually international orientation from its earliest years and embedded an enduring tradition of training students who would split their academic lives between China and the West.
In 1925 Tsinghua launched its own four year undergraduate program and opened a research institute focused on Chinese studies. In 1928, under the Republic of China, the school was renamed National Tsinghua University. During the Second Sino Japanese War (1937 to 1945) the university was forced south, where it merged with Peking University and Nankai University to form National Southwestern Associated University in Kunming, an institution that became legendary for producing many of the scientists who would later lead China's nuclear, aerospace, and computing programs. After the war Tsinghua returned to Beijing.
Following the founding of the People's Republic of China in 1949, the university was reorganized into a polytechnic with a heavy focus on engineering. This restructuring shaped the modern identity of the school: while Peking University retained the humanities and sciences flagship role, Tsinghua became the country's leading engineering and applied sciences university. The reform era of the 1980s and 1990s saw Tsinghua expand back into management, economics, law, and the natural sciences, but the engineering DNA remained, and it shaped the way the institution would later approach computing and AI.
Tsinghua is consistently ranked among the world's leading universities. In the 2026 Times Higher Education World University Rankings the institution placed 12th globally and first in Asia. In the 2026 QS World University Rankings it placed 17th. The Academic Ranking of World Universities placed Tsinghua 18th worldwide in 2025. In the 2025 Nature Index, which measures contributions to high impact natural science publications, Tsinghua ranked 6th globally and first among Chinese universities.
For computer science specifically, Tsinghua is the highest ranked Chinese institution and is regularly placed in the global top twenty. Various subject specific rankings and citation indices consistently identify it as the top Chinese university for AI research, although Peking University, Shanghai Jiao Tong University, and the Chinese Academy of Sciences are close peers. The university enrolls roughly 50,000 students across undergraduate, master's, and doctoral programs, with computer science and electronic engineering being among the largest and most selective majors.
Tsinghua's engagement with computing dates back to 1958, when the university established the Department of Automatic Control. The Computer Science major was carved out of the Radio Department and folded into Automatic Control. By the 1970s and 1980s, with strong central government investment in computing as a strategic technology, the department had become an independent unit and began building up faculty in operating systems, computer architecture, networking, and what was then called pattern recognition.
The modern Department of Computer Science and Technology comprises six institutes:
The Institute of Artificial Intelligence is the formal hub for AI work inside the department, although the AI agenda has expanded well beyond it. The Knowledge Engineering Group (KEG), the Natural Language Processing Group (THUNLP), the Machine Learning Group, and the Statistical AI Lab all operate as semi independent research clusters under the broader department, each with its own faculty, students, and industrial partnerships. Beyond the department itself, the university operates several cross disciplinary AI institutes that pull faculty from electronics, biomedical engineering, and the social sciences.
The single most important AI talent pipeline at Tsinghua is the Institute for Interdisciplinary Information Sciences (IIIS), led by Professor Andrew Yao, the only Chinese born winner of the Turing Award. Yao received the Turing Award in 2000 "in recognition of his fundamental contributions to the theory of computation, including the complexity based theory of pseudorandom number generation, cryptography, and communication complexity." He had previously held faculty positions at MIT, Stanford, UC Berkeley, and Princeton before leaving Princeton in 2004 to return to China full time. The decision drew international attention and was widely seen as a turning point in the global flow of computer science talent.
In 2005 Yao founded the Yao Class (姚班), an elite undergraduate program in computer science. The Yao Class admits a small cohort each year, around 40 to 50 students, drawn from the top performers in the gaokao college entrance examination. Yao personally designed the curriculum based on his teaching experience at MIT, Stanford, and Princeton. Courses are taught largely in English and emphasize theoretical foundations, including algorithms, complexity, cryptography, and quantum information, alongside applied work in machine learning and AI. The program now offers three specialized tracks: computer science, artificial intelligence, and quantum information.
In 2011 Yao established IIIS as the institutional home for the Yao Class along with three sister programs: the Yao Class Graduate Program, the Quantum Information Center, and the Center for Theoretical Computer Science. Today IIIS supports research that spans computer science, quantum information, and AI, with extensive interdisciplinary work in economics, manufacturing, healthcare, and finance. The Yao Class has been recognized with China's top national higher education achievement award and is widely viewed as the most academically demanding undergraduate program in the country.
The alumni network of the Yao Class reads like a who's who of the modern Chinese AI industry. Wang He, founder of Beijing Galaxy General Robot, is a Yao Class graduate. Yao Class alumni populate the founding research teams of Zhipu AI, Moonshot AI, MiniMax, and many other firms in the so called Six Tigers cohort, as well as international labs including DeepMind, OpenAI, Google Brain, and Anthropic, which produces Claude.
In 2020 Tsinghua launched the Institute for AI Industry Research (AIR, 智能产业研究院), a translation oriented institute charged with applying AI to specific industrial domains. AIR is led by founding dean Zhang Yaqin, a Chair Professor of AI Science at Tsinghua and one of the most senior figures in the Chinese tech industry. Zhang previously served as President of Baidu from 2014 to 2019, where he oversaw cloud computing, AI silicon, autonomous driving, and industrial AI initiatives. Before Baidu, he spent sixteen years at Microsoft, including stints as Managing Director of Microsoft Research Asia and Chairman of Microsoft China. He is an elected member of the Chinese Academy of Sciences, the American Academy of Arts and Sciences, the National Academy of Inventors, and the Australia National Academy of Engineering.
AIR concentrates on three application areas: autonomous driving, AI plus green computing, and AI plus life sciences. It operates somewhat differently from the traditional academic departments by emphasizing industrial co projects, joint hiring with companies, and applied deliverables rather than pure publications. The institute has become a magnet for senior researchers returning from industry, and it has played an important role in connecting frontier academic work to commercial deployment in the Chinese auto industry, biotech, and energy sectors.
The Knowledge Engineering Group (KEG) is one of the oldest and most influential research labs at Tsinghua. Led by Professor Tang Jie, a WeBank Chair Professor and Fellow of ACM, AAAI, and IEEE, KEG has worked on knowledge graphs, data mining, social networks, and machine learning since the early 2000s. Tang's work on the AMiner academic search engine, which indexes hundreds of millions of researchers and publications, won the SIGKDD Test of Time Award in 2024.
In the early 2020s KEG pivoted decisively into large language models. The lab released a sequence of pretrained models in the GLM (General Language Model) family. The breakthrough release was GLM-130B, an open bilingual English and Chinese pretrained model with 130 billion parameters, published as a paper at ICLR 2023. On English benchmarks GLM-130B outperformed GPT-3 175B davinci by 5.0 percent on LAMBADA and slightly outperformed it on MMLU. On Chinese benchmarks it significantly outperformed Baidu's ERNIE TITAN 3.0 260B on standard CLUE and FewCLUE evaluations. The training itself was sponsored by Zhipu AI, which provided 96 A100 servers for the run.
The GLM line evolved into the chat oriented ChatGLM models, which became among the most widely downloaded open source large language models from China. The progression went from ChatGLM (March 2023) through ChatGLM2 and ChatGLM3 to GLM-4 All Tools, an agentic model released in 2024 that integrated tool use, code execution, and web browsing. The ChatGLM series has been downloaded over 30 million times globally on Hugging Face and other platforms, making it one of the most influential open source model lineages of the early LLM era.
KEG has also produced multimodal models including CogView2 for text to image generation and CogVideo for text to video, both built on hierarchical transformer architectures. Co director Li Juanzi, also a Tsinghua professor, has incubated multiple AI startups out of the lab, including DeepLang AI and contributing personnel and methods to Moonshot AI.
The Natural Language Processing Group at Tsinghua University (THUNLP) is the country's oldest NLP lab, founded in the 1970s under Professor Changning Huang to work on Chinese information processing. It is now led by Professor Sun Maosong, a full professor in the Department of Computer Science and Technology and executive deputy director of the Institute for Artificial Intelligence at Tsinghua. THUNLP's faculty includes Professor Yang Liu and Associate Professor Zhiyuan Liu, both of whom have produced widely cited work on knowledge graphs, neural machine translation, and pretrained models.
THUNLP's open source toolkits, including THUMT for machine translation and OpenKE for knowledge embeddings, have been used by hundreds of research groups worldwide. Sun Maosong serves as chief scientist of DeepLang AI, an NLP startup founded by Tsinghua alumni, and his students have founded or joined many of the leading Chinese LLM teams.
Professor Zhu Jun leads Tsinghua's Statistical AI and Bayesian methods program. Zhu is an ACM, IEEE, and AAAI Fellow whose research focuses on probabilistic machine learning, adversarial robustness, Bayesian methods, deep learning, and reinforcement learning. He directs the Basic Theory Research Center at the Tsinghua Institute for AI and held an adjunct faculty appointment at Carnegie Mellon University from 2015 to 2018. In 2017 his team released ZhuSuan, a Python probabilistic programming library that integrated Bayesian methods with deep learning and supported efficient multi machine, multi GPU training. Zhu cofounded RealAI, a startup focused on AI safety and adversarial robustness, in 2018.
| Faculty | Role | Field | Notable contributions |
|---|---|---|---|
| Andrew Yao | Dean, IIIS | Theoretical CS, AI | 2000 Turing Award, founder of Yao Class and IIIS |
| Tang Jie | WeBank Chair Professor, KEG | NLP, knowledge graphs, LLMs | GLM-130B, ChatGLM, AMiner, BAAI Wu Dao lead |
| Sun Maosong | Professor, THUNLP | NLP, computational linguistics | Founder of modern THUNLP, chief scientist of DeepLang AI |
| Zhu Jun | Professor, Department of CS | Bayesian methods, deep generative models | ZhuSuan library, cofounder of RealAI |
| Li Juanzi | Professor, KEG | Knowledge engineering, LLMs | Cofounder of Zhipu AI, ChatGLM contributor |
| Zhang Yaqin | Chair Professor, AIR Dean | Industrial AI, autonomous driving | Founding dean of AIR, former Baidu President |
| Yang Liu | Professor, THUNLP | Machine translation | THUMT toolkit, neural translation research |
| Zhiyuan Liu | Associate Professor, THUNLP | Pretrained models, knowledge embeddings | OpenKE, network embeddings |
| Xiaolin Hu | Associate Professor, Department of CS | Computer vision, computational neuroscience | Brain inspired vision models |
| Yuanchun Li | Assistant Professor, AIR | Mobile intelligence, on device AI | On device LLMs, formerly at MSRA |
The density of Tsinghua linked AI startups is unusual even by global elite university standards. According to a 2023 industry survey by Zhidongxi covering 11 leading domestic LLM enterprises, 17 of their founders had Tsinghua ties. By the mid 2020s, that share had grown rather than shrunk: of the so called Six Tigers, the cohort of leading Chinese LLM startups that became unicorns during 2023 and 2024, four had primary Tsinghua origins.
| Company | Founded | Tsinghua connection | Focus |
|---|---|---|---|
| Zhipu AI (now Z.ai) | 2019 | Spun out of Tsinghua KEG; cofounded by professors Tang Jie and Li Juanzi | Foundation models, ChatGLM, GLM-4 |
| Moonshot AI | March 2023 | Founded by Tsinghua alumni Yang Zhilin, Zhou Xinyu, and Wu Yuxin | Long context LLMs, Kimi chatbot |
| Baichuan Intelligence | March 2023 | Founder Wang Xiaochuan was a Tsinghua undergraduate; many engineers from Tsinghua | Bilingual foundation models |
| MiniMax | December 2021 | Cofounders include Tsinghua alumni; close THU research ties | Multimodal foundation models, video and voice |
| Beijing Galaxy General Robot (Galbot) | May 2023 | Founder Wang He holds a Tsinghua bachelor's degree | Embodied AI, humanoid robotics |
| DeepLang AI | March 2022 | Founded by Tsinghua alumni Qi Fanchao and Li Xiaoxiang; chief scientist Sun Maosong of THUNLP | Chinese language NLP, dictionaries, classical poetry |
| Shengshu Technology | 2023 | Founded by Tsinghua professor Zhu Jun and students | Multimodal generative AI, Vidu video model |
| RealAI | 2018 | Founded by Zhu Jun's group at Tsinghua | AI safety, adversarial robustness |
| Galaxea AI | 2023 | Founded by Tsinghua alumni and faculty | Humanoid robotics, embodied AI |
| Robotera | 2023 | Tsinghua alumni founders | Humanoid robots |
| Noetix Robotics | 2023 | Tsinghua alumni founders | Humanoid robots |
| Booster Robotics | 2023 | Tsinghua alumni founders | Humanoid robots |
The rate of company formation accelerated dramatically after the release of ChatGPT in November 2022. In 2023 alone, dozens of foundation model and embodied AI startups were founded by Tsinghua faculty or alumni, often staffed by recent Yao Class graduates and KEG students. By 2024 venture capital firms were openly tracking the "Tsinghua camp" as a distinct investment thesis, with some funds maintaining dedicated portfolios of Tsinghua linked startups.
Tsinghua's AI research is spread across multiple labs and institutes, some inside the Department of Computer Science and Technology and others as cross disciplinary institutes that report directly to the university leadership. The most important are summarized below.
| Lab or institute | Founded | Lead | Focus |
|---|---|---|---|
| Institute for Interdisciplinary Information Sciences (IIIS) | 2011 | Andrew Yao | Theoretical CS, AI, quantum information; home of the Yao Class |
| Institute for AI Industry Research (AIR) | 2020 | Zhang Yaqin | Industrial AI, autonomous driving, AI for life sciences |
| Knowledge Engineering Group (KEG) | early 2000s | Tang Jie, Li Juanzi | Knowledge graphs, LLMs, GLM and ChatGLM |
| Natural Language Processing Lab (THUNLP) | 1970s | Sun Maosong | NLP, machine translation, pretrained models |
| Institute for Artificial Intelligence | 2018 | rotates | University level AI institute, coordinates cross departmental AI work |
| Statistical AI Lab | 2010s | Zhu Jun | Bayesian deep learning, generative models, AI safety |
| Institute for AI International Governance (I-AIIG) | 2020 | Xue Lan | AI policy, governance, international cooperation |
| Tsinghua Berkeley Shenzhen Institute (TBSI) | 2014 | joint with UC Berkeley | AI, machine learning, big data, cybersecurity |
| Tsinghua x lab | 2013 | rotating | Student entrepreneurship across 14 schools |
| Tsinghua Science Park / TusStar | 1994 / 2001 | TusHoldings | Deep tech and AI incubation |
Tsinghua faculty have played central roles in the Beijing Academy of Artificial Intelligence (BAAI), a non profit research lab founded in 2018 with funding from the Beijing municipal government and major Chinese tech companies. BAAI's flagship achievement of the early LLM era was the Wu Dao family of models. Wu Dao 2.0, unveiled in May 2021, was a 1.75 trillion parameter multimodal pretrained model and was at the time the largest such model publicly disclosed anywhere in the world.
The Wu Dao 2.0 effort was led by BAAI Research Academic Vice President Tang Jie, who was simultaneously a Tsinghua professor in KEG. Roughly 100 AI scientists from Tsinghua, Peking University, Renmin University of China, and the Chinese Academy of Sciences contributed to the project. BAAI also unveiled Hua Zhibing, a virtual student powered by Wu Dao that was officially enrolled in Tsinghua's Department of Computer Science and Technology. The Wu Dao project served as a kind of dress rehearsal for the wave of Chinese LLM activity that followed ChatGPT.
Tsinghua faculty and students have produced or co produced many of the benchmarks now used to evaluate Chinese and bilingual large language models.
Tsinghua's open source releases on Hugging Face under the THUDM (Tsinghua University Data Mining) organization include the entire ChatGLM family, GLM-130B, CogView, CogVideo, and dozens of other model and dataset releases. As of 2024 THUDM was among the most followed Chinese organizations on the platform.
Tsinghua maintains an extensive network of international research partnerships in AI and related fields. By the end of 2024 the university had signed strategic scientific research cooperation agreements with 24 universities and research institutions worldwide.
The most prominent AI focused partnership is the Tsinghua Berkeley Shenzhen Institute (TBSI), a joint institute established in 2014 with the University of California, Berkeley, and located in Shenzhen. TBSI focuses on artificial intelligence, machine learning, big data, and cybersecurity, and runs joint master's and doctoral programs with student exchange visits between Beijing, Shenzhen, and Berkeley. The institute also serves as a major recruiting ground for Chinese AI startups operating in the Pearl River Delta.
The Schwarzman Scholars program, founded by American financier Stephen Schwarzman and headquartered at Schwarzman College on the Tsinghua campus, brings around 150 international graduate students to Beijing each year for a fully funded master's degree in global affairs. Although not exclusively focused on AI, the program has become a key node for AI policy work and has produced alumni working on AI governance, autonomous systems regulation, and US China technology policy. The program has been described as a deliberate counterpart to the Rhodes Scholarship.
Tsinghua's faculty maintain extensive academic exchanges with MIT, Stanford, Princeton, Carnegie Mellon, Cambridge, Oxford, ETH Zurich, and the University of Tokyo. Many senior Tsinghua AI faculty hold or have held visiting appointments at these institutions, and the university sends large cohorts of graduate students for one or two year research stays abroad. The Institute for AI International Governance, led by Professor Xue Lan, runs an annual international AI governance forum that has brought together policymakers, researchers, and industry leaders from China, the United States, the European Union, and other regions.
The broader Tsinghua ecosystem extends well beyond the university's gates. Tsinghua Science Park, established in 1994, is one of the largest university affiliated technology parks in the world and houses thousands of high technology firms in Beijing's Zhongguancun district. TusHoldings and its incubator arm TusStar operate a global network of incubators and accelerators that have backed many Tsinghua linked AI startups. Tsinghua x lab, founded in 2013, links 14 schools and departments across the university and provides early stage support for student founders.
By 2022, Tsinghua alumni had built an entrepreneurial and investment network of approximately 1,632 entrepreneurs and 633 investors, creating a self reinforcing financing pipeline for new Tsinghua linked ventures. This dense network is sometimes referred to in Chinese media as the "Tsinghua camp" (清华系) and is regularly compared to the Stanford or MIT alumni networks in Silicon Valley.
The ecosystem is further reinforced by the geographic concentration of Chinese AI activity in Beijing's Haidian district, where Tsinghua, Peking University, the Chinese Academy of Sciences, and BAAI all sit within a few square kilometers, surrounded by the offices of Baidu, ByteDance, Microsoft Research Asia, Zhipu AI, Moonshot AI, and most of the other firms in the China AI industry.
Tsinghua's central role in Chinese AI has not been without controversy. The university has been the subject of US export control discussions because some of its labs work in fields with dual use military and civilian applications. Some Western collaborators have ended or scaled back joint projects in response to US Department of Commerce entity list designations affecting affiliated Chinese organizations, although Tsinghua itself has not been placed on the entity list as of 2026.
The outsized concentration of Chinese AI talent at Tsinghua has also prompted concern within China about over reliance on a single institution. Critics argue that a healthier national AI ecosystem would distribute talent more evenly across regional universities. A recurring theme in Chinese tech press coverage, including analyses by Caixin and Sixth Tone, is whether the Tsinghua dominated AI scene risks becoming a closed network that crowds out founders without elite credentials. Defenders point out that Tsinghua's openness to external talent, including its active recruitment of returnees from US PhD programs and industry, has made the network more permeable than the equivalent networks at older Chinese institutions.