Yang Zhilin

Long before founding Moonshot AI, Yang Zhilin co-authored influential work on long-context language modelling. Kimi K2 brought that research lineage into a trillion-parameter mixture-of-experts system with released checkpoints, while independent evaluation placed its progress in a measured international context.

Yang Zhilin studied computer science at Tsinghua University and completed doctoral work at Carnegie Mellon University. He co-authored Transformer-XL, published at ACL in 2019, and contributed to XLNet, research that addressed context length and dependency modelling before long-context systems became a major commercial category. He later co-founded Moonshot AI in Beijing and became its chief executive. The company’s consumer assistant, Kimi, made long-context interaction central to its early identity. Adoption totals are principally company-reported and should not be confused with independently audited use. In July 2025, Yang was among the authors of the Kimi K2 technical report. The system used a mixture-of-experts architecture with one trillion total parameters and 32 billion active parameters, trained on 15.5 trillion tokens. The report also described MuonClip, an optimisation approach intended to stabilise large-scale training, and released model checkpoints for external use. These were team achievements, not Yang’s work alone. An independent December 2025 evaluation by a national standards institute described K2 Thinking as the most capable publicly available Chinese model it had evaluated at release, while finding that it remained behind leading United States systems on several dimensions. That balanced result supports Yang’s inclusion more strongly than unqualified company benchmarking would.