Yang Zhilin

Overview

Yang Zhilin is a Chinese AI researcher and entrepreneur. He is a co-founder and CEO of Moonshot AI, the Beijing company behind the Kimi assistant.^[1][2] His career runs from machine-learning research at Tsinghua and Carnegie Mellon, through the research labs of Google and Meta, to founding and running a company that trains and publishes its own large models.^[1][2] Research and business have run alongside each other throughout.

Education and early research

Yang entered Tsinghua University in 2011. He first studied thermodynamics and later switched to computer science, joining the lab of Tang Jie.^[2] Switching fields meant starting again on the basics of a different subject. From Tsinghua he went to Carnegie Mellon University for a PhD under Ruslan Salakhutdinov and William Cohen, and finished the degree in four years.^[2] During the PhD he interned at Google Brain and at Meta AI Research.^[2]

Career

Before founding his own company, Yang worked at Google Brain, Meta and Huawei; at Huawei he was involved with the PanGu family of models.^[1] PanGu is Huawei's own family of models. On this period, Caixin's profile of Yang describes Huawei Cloud's 2021 work on the Pangu NLP model with Recurrent AI as a partnership rather than as a period of employment at Huawei.^[2]

Main contributions

Yang had co-founded a startup called Recurrent AI. In early 2023 he and Zhang Yutao left that company to create Moonshot AI.^[2] The new company was founded in Beijing in March 2023 with two Tsinghua schoolfriends, Zhou Xinyu and Wu Yuxin, and Yang took the CEO role.^[1][2]

Under Yang's leadership Moonshot released the Kimi assistant in October 2023.^[1] An assistant of this kind is a chat-style program: the user asks a question in ordinary language, and a large model trained on a great deal of text produces the reply. A product and a model are two separate lines of release — the assistant is aimed at end users, the model at developers.

The open-weight Kimi K2 followed in July 2025, and Kimi K3 in July 2026.^[1][3] "Open weight" means the trained parameters are published so that others can download and run the model in their own environment, rather than reaching it only through a company's interface; whoever has the weights can adapt and deploy the model, but needs the computing power and engineering to run it. A closed service, by contrast, keeps the weights inside the company, and users can only call the model on the provider's servers.

Impact

Under Yang, Moonshot's public releases have fallen into two categories: a consumer-facing assistant and open-weight models.^[1][3] The company was started by three people who had studied at Tsinghua — Yang and two schoolfriends — and it came straight after he left the previous startup he had co-founded.^[1][2]