Yao Xingcheng, formerly responsible for pretraining at Tencent's Hunyuan large language model team, has recently joined North American startup Thinking Machines Lab, according to a report by Lei Feng Wang. The closely watched young AI researcher ultimately chose to join an overseas AI lab rather than stay at a major Chinese tech company.
According to people familiar with the matter, the compensation package Thinking Machines Lab offered Yao was noticeably higher than his pay at Tencent, where his annual package was in the range of tens of millions of RMB. He left Tencent in July of this year, where his main responsibility was pretraining work on Tencent's Hunyuan large language model.
Why Yao Xingcheng chose to join Thinking Machines Lab
Notably, Yao had previously been reported as heading to Meta, but ultimately chose Thinking Machines Lab, the AI company founded by OpenAI's former chief technology officer Mira Murati. The decision has given the industry a fresh sense of the high-profile startup's pull.
In terms of background, Yao Xingcheng and Yao Shunyu, another well-known researcher in China's large model field, both came out of Tsinghua University's Yao Class and are classmates under the same mentor. The Yao Class has long supplied a steady stream of top talent to China's large model industry, and a portion of that cohort has now begun moving overseas.
For China's tech giants, top-tier pretraining talent is becoming an increasingly scarce resource. Lei Feng Wang previously reported exclusively on Chinese majors poaching pretraining talent from Gemini; now Tencent's own pretraining talent is also flowing toward overseas labs. Industry observers note that pretraining is the bedrock of large model capability, and the comings and goings of core researchers often directly shape a company's long-term competitiveness in foundation models. The fight for top pretraining talent has extended from China to the global stage.