Tencent Appoints Former OpenAI Researcher Tian Yonglong to Lead Hunyuan Multimodal
Tencent has appointed former OpenAI researcher Tian Yonglong as head of Hunyuan multimodal, reporting directly to Yao Shunyu. The company's Hunyuan Hy4 preview model recorded 14.7 trillion weekly tokens, ranking first globally. China's total weekly AI model usage reached 56.72 trillion tokens, surpassing the US for the 19th consecutive week.
On September 9, Tian Yonglong, a former researcher at OpenAI, joined Tencent's CEO/President's Office as head of Hunyuan multimodal, reporting directly to Yao Shunyu, Tencent's head of large language models. On July 8, Tian announced that he had recently joined Tencent's large language model department and would participate in research and development related to visual language models (VLMs). Tian's core research areas at OpenAI included computer vision, visual representation learning, and generative models.
Before joining OpenAI, Tian spent many years at Google. In late 2022, he became a senior research scientist at Google Research in Cambridge, and in May 2024, he moved to Google DeepMind. Tian had previously worked alongside Yao Shunyu, now Tencent's chief AI scientist, during their time at OpenAI. On December 17, 2025, Tencent announced an upgrade to its large model research architecture, establishing the AIInfra, AIData, and Data Computing Platform departments, and appointing Yao Shunyu, a former senior researcher at OpenAI, as chief AI scientist in the CEO/President's Office (reporting to Liu Chiping), while also serving as head of both the AIInfra and Large Language Model departments (reporting to Lu Shan).
In the week ending September 6, global AI large model usage totaled 115 trillion tokens, with Chinese models accounting for 56.72 trillion tokens, surpassing the US for the 19th consecutive week. Among them, Tencent's Hunyuan Hy4 preview ranked first globally with 14.7 trillion weekly tokens, a 379% increase from the previous week. GPT-5.6 Luna ranked second, and Zhipu's GLM-5.3 Flash rose to third. Four of the top five models globally were Chinese. Hy4 preview, released and open-sourced by Tencent on August 28, is a next-generation MoE flagship model with 770 billion total parameters, 49 billion active parameters, and a 1 million token context window.