Tsinghua KEG

The Knowledge Engineering Group (KEG) is an academic research group within the Department of Computer Science and Technology at Tsinghua University in Beijing, China. KEG was established as a research group focused on knowledge graphs, academic data mining, and other areas, and is led by Jie Tang, a senior figure in Chinese computer-science research and a Fellow of the ACM, AAAI, and IEEE. KEG is institutionally significant beyond its standalone academic research output because the group produced the foundational GLM (General Language Model) and CogView (text-to-image) research that became the basis for Z.ai (formerly Zhipu AI), the commercial spinout that completed a Hong Kong IPO in January 2026 as the first pure-play foundation-model developer to go public globally. Tsinghua KEG continues to operate as an academic research group alongside the commercial Z.ai entity and is the principal Chinese academic locus for foundation-model and knowledge-engineering research.

At a glance

• Founded: Approximately 2012 in its modern form as the Knowledge Engineering Group within Tsinghua University's Department of Computer Science and Technology. Earlier predecessor research at Tsinghua dates to the 2000s and earlier.
• Status: Academic research group within Tsinghua University's Department of Computer Science and Technology. Operates as part of the academic research-and-education structure of one of China's most prestigious research universities.
• Funding: Tsinghua University institutional support, Chinese government research grants (NSFC and other agencies), and corporate-and-foundation collaboration support. KEG also has connection to the commercial Z.ai funding base through founder participation.
• CEO: Jie Tang, Professor, Department of Computer Science and Technology at Tsinghua University, and head of the Knowledge Engineering Group. Director of the Foundation Model Research Center at Tsinghua's AI Institute. ACM Fellow, AAAI Fellow, IEEE Fellow. Co-founder of Z.ai (formerly Zhipu AI).
• Other notable leadership: Juanzi Li, Tsinghua professor and KEG senior researcher. Co-founder of Z.ai. Senior research-program leadership across the affiliated KEG faculty cohort.
• Open weights: Yes. KEG research outputs are released openly through Hugging Face under the THUDM (Tsinghua University Data Mining) organization. Commercial-grade variants are distributed through the Z.ai commercial entity.
• Flagship outputs: GLM-130B (open-weights bilingual Chinese-and-English foundation model, August 2022), ChatGLM-6B (open-weights conversational AI model, March 2023), CogView (text-to-image model, May 2021), CogVideo (video-generation model, May 2022), and the AMiner academic search platform.

Origins

Tsinghua University's Knowledge Engineering Group (KEG) was established in its modern form approximately 2012 within the Department of Computer Science and Technology, building on earlier knowledge-engineering, knowledge-graph, and academic data-mining research at Tsinghua. The group's founding research direction emphasized academic-publication mining, social-network analysis among researchers, and other applications of computer-science research methods to academic information.

Jie Tang, the KEG leader, established the AMiner platform (an academic search and analytics service) as one of the group's principal early research artifacts. AMiner became one of the most-used academic-search platforms in Chinese academia and contributed public-domain data on academic publications, citations, and researcher networks.

The 2018 to 2022 period saw KEG expand into foundation-model research alongside the broader transformation of Chinese AI research. Jie Tang and Juanzi Li led the development of the GLM (General Language Model) pre-training architecture in late 2020, with the GLM-10B model trained in 2021. The GLM-130B model in August 2022 was a foundation-model release: 130 billion parameters, bilingual Chinese-and-English training, and open-weights distribution through Hugging Face under the THUDM organization. The CogView text-to-image model in May 2021 was one of the earliest Chinese text-to-image research systems with competitive quality, and the CogVideo video-generation model in May 2022 extended the multimodal research line.

In June 2019, Tang Jie and Juanzi Li had founded Zhipu AI as a commercial spinout from KEG, with the company initially incubated within Tsinghua's Science Park in Beijing. The Zhipu AI commercial entity has subsequently become Z.ai, completed a $558 million Hong Kong IPO in January 2026, and is now publicly listed (HKEX: 2513). KEG continues to operate as an academic research group at Tsinghua alongside the commercial Z.ai entity, with personnel and research overlap through Jie Tang's and Juanzi Li's continued affiliation with both organizations.

The 2023 to 2026 period has seen KEG continue research output across foundation-model research, multimodal AI, knowledge engineering, and other areas. The ChatGLM-6B release in March 2023 was an early Chinese open-weights conversational AI model that established Tsinghua's open-source-AI credibility. Subsequent THUDM open-weights releases have continued through Hugging Face. The relationship between the academic KEG group and the commercial Z.ai entity has matured into a structurally distinctive academic-and-commercial dual structure that is unusual among contemporary AI research organizations.

Mission and strategy

Tsinghua KEG's stated mission is to advance research in knowledge engineering, foundation-model AI, knowledge graphs, and other areas through academic research, open-source releases, and education. The group operates as an academic research unit at Tsinghua, with senior faculty leading research programs that produce academic publication, open-source releases, and other contributions.

The strategy combines four threads. First, fundamental research on foundation models, knowledge engineering, and other areas, with senior-faculty-led research programs producing academic publication output. Second, open-source release of research-stage and earlier-generation foundation-model variants through Hugging Face under the THUDM organization. Third, talent development through Tsinghua's graduate and undergraduate programs, with KEG producing PhD-level researchers who subsequently move to commercial AI labs (KEG alumni representation across Chinese AI organizations) and academic positions. Fourth, the Z.ai commercial spinout structure, providing a structural channel for commercial productization of KEG research outputs.

The competitive premise is that academic AI research at Tsinghua, particularly with the foundation-model research depth KEG has assembled, can produce structurally important contributions to the broader AI ecosystem. The dual academic-and-commercial structure with Z.ai is unusual and provides a model for how academic AI research can complement and feed commercial AI development.

Models and products

Tsinghua KEG is a research group rather than a model-development organization in the conventional commercial sense. The group's outputs include:

• GLM-130B. Released August 2022. 130-billion-parameter open-weights bilingual Chinese-and-English foundation model. The flagship academic foundation-model contribution from KEG before the Z.ai commercial productization.
• ChatGLM-6B. Released March 2023. Open-weights conversational AI model. The principal early Chinese open-weights conversational AI model.
• CogView and CogVideo. Text-to-image (May 2021) and video-generation (May 2022) research models. Foundational Chinese multimodal research output.
• GLM family open-weights releases. Subsequent GLM-style open-weights releases through THUDM on Hugging Face.
• AMiner. Academic search and mining platform. Long-running KEG project that anchors the group's knowledge-engineering research.
• Knowledge graph research. Research output on knowledge graphs and their integration with language models.
• Foundation Model Research Center at Tsinghua's AI Institute. Senior-faculty-led research center directed by Jie Tang.

The principal distribution channel is academic-paper publication, GitHub for training code and infrastructure, and Hugging Face for model and dataset releases through the THUDM organization. Commercial-grade GLM variants are distributed through the Z.ai commercial entity rather than through the academic KEG channel.

Benchmarks and standing

Tsinghua KEG is a research group that contributes to the AI research community rather than competing on capability benchmarks. The group's research-community contributions are measured through academic publication output, citation impact, the GLM and CogView research line, and the Z.ai commercial spinout that has subsequently become a publicly listed company.

KEG is consistently ranked among the most-productive Chinese academic AI research groups by international and domestic academic rankings. Jie Tang is one of the most-cited Chinese computer-science researchers and holds Fellow status in the ACM, AAAI, and IEEE.

KEG's standing in the global AI research community is anchored on the foundational GLM and CogView research, the Z.ai commercial validation, the open-weights release legacy, and the senior research-faculty depth.

Leadership

As of April 2026, Tsinghua KEG's senior leadership includes:

• Jie Tang, Professor in the Department of Computer Science and Technology at Tsinghua University. Head of KEG. Director of the Foundation Model Research Center at Tsinghua's AI Institute. ACM Fellow, AAAI Fellow, IEEE Fellow. Co-founder of Z.ai. Public face for both the academic KEG group and the commercial Z.ai entity.
• Juanzi Li, Tsinghua professor and KEG senior researcher. Co-founder of Z.ai. KEG faculty member with research output on knowledge graphs and foundation-model research.
• Senior research-faculty cohort within the Knowledge Engineering Group at Tsinghua, with multiple senior researchers and PhD students contributing to the group's research output.

The group's structure includes the dual academic-and-commercial relationship with Z.ai, with Jie Tang and Juanzi Li participating in both organizations. The post-2019 commercial Z.ai entity provides commercial productization of foundation-model research while the academic KEG group continues to produce research output.

Funding and backers

Tsinghua KEG's capital structure is the academic-research-group model funded through Tsinghua University institutional support, Chinese government research grants (NSFC and other agencies), and corporate-and-foundation collaboration support. Specific cumulative funding figures combine the group's institutional budget with the research-grant portfolios of individual affiliated faculty.

The commercial Z.ai entity provides a structurally distinctive complementary funding mechanism. Z.ai's pre-IPO funding rounds (approximately $1.4 billion) and the January 2026 Hong Kong IPO ($558 million raised at debut) provide commercial-scale funding for foundation-model research that complements the academic KEG group's institutional support.

The dual academic-and-commercial structure produces unusual capability for sustained foundation-model research investment, with KEG's academic budget and Z.ai's commercial funding providing complementary resources.

Industry position

Tsinghua KEG occupies a structurally distinctive position in the global AI research landscape. The combination of the foundational GLM and CogView research outputs, the senior-faculty leadership of Jie Tang and Juanzi Li, the Z.ai commercial spinout (now publicly listed at HKEX), and the dual academic-and-commercial structure produces a profile that no other academic AI research group matches at the same combination of attributes.

Industry coverage frequently characterizes Tsinghua KEG as the principal Chinese academic AI research group with foundation-model research depth, with the Z.ai spinout as the most successful Chinese academic-AI commercial transition in the contemporary period. Tsinghua's broader research-and-engineering reputation amplifies the group's research-community standing.

Strategic risks include intensifying competition for AI research talent from Chinese commercial AI labs (DeepSeek, Alibaba Qwen, Moonshot AI, Z.ai itself, and other organizations), the open question of whether academic AI research can keep pace with commercial frontier-model investment, and the broader Chinese AI policy environment. Strategic strengths include the Tsinghua academic prestige, the senior-faculty research depth, the Z.ai commercial spinout structural support, and the open-source-AI release legacy through THUDM.

Competitive landscape

Tsinghua KEG collaborates with and complements rather than directly competes with most other AI organizations:

• Z.ai (formerly Zhipu AI). Direct commercial spinout from KEG. Personnel and research overlap through Jie Tang and Juanzi Li's continued affiliation with both organizations.
• Peking University, Shanghai Jiao Tong University, USTC. Peer Chinese academic AI research institutions.
• DeepSeek, Alibaba Qwen, Moonshot AI, MiniMax, StepFun, 01.AI, ByteDance Seed, Tencent Hunyuan, Baidu. Chinese commercial AI labs that recruit KEG graduates and collaborate on research projects.
• Stanford HAI / CRFM, Berkeley BAIR, MIT CSAIL, CMU SCS, KAIST. International academic AI research peers.
• Allen Institute for AI, Hugging Face, EleutherAI, LAION, BigScience, MILA, Nous Research. Open-AI-research peer organizations.
• Beijing Academy of Artificial Intelligence (BAAI). Chinese AI research organization with overlap in foundation-model research.

Outlook

Several open questions affect Tsinghua KEG's trajectory in 2026 and 2027:

• The continued evolution of the dual academic-and-commercial structure with Z.ai, including the balance of research-and-personnel allocation between the academic group and the commercial entity.
• Continued open-weights research-stage foundation-model releases through THUDM.
• Senior research-talent recruitment and retention against Chinese commercial AI labs.
• The group's role in shaping Chinese AI policy through Tsinghua's broader engagement with Chinese government research-funding agencies.
• Continued international collaboration with US and European academic AI research peers.
• The development of Tsinghua's broader AI research initiatives including the Foundation Model Research Center and other organizational structures.
• The continued contribution of the AMiner academic-search platform and other knowledge-engineering research outputs.

Sources

• Knowledge Engineering Group (KEG) at Tsinghua University. Group overview reference.
• Jie Tang's homepage. KEG leader profile.
• THUDM on Hugging Face. Open-weights model distribution.
• GitHub: THUDM/GLM-130B. GLM-130B model release.
• GitHub: THUDM/ChatGLM-6B. ChatGLM-6B release.
• GitHub: THUDM/CogView. CogView text-to-image research.
• Wikipedia: Z.ai. Commercial-spinout reference for KEG context.