36Kr Reports Tencent Hunyuan Multimodal Lead Hu Han Resigned
36Kr reported July 23 that Hu Han, head of multimodal understanding at Tencent Hunyuan, tendered his resignation to launch a startup. Former OpenAI researcher Tian Yonglong is expected to take over vision-language-model R&D, while Hu's former group may shift toward world-model research. Tencent had not responded, and KrASIA's July 27 article is an adaptation of the same 36Kr report rather than independent corroboration.
36Kr reported on July 23 that Hu Han, who led multimodal understanding at Tencent's Hunyuan unit, tendered his resignation and plans to start a company. The outlet said former OpenAI researcher Tian Yonglong will take over research and development for vision-language models and report to Yao Shunyu, head of Tencent's large-language-model department.
Tencent had not responded to the report by publication time. KrASIA published an English adaptation on July 27, but it explicitly identifies 36Kr as the original source, so the two articles are one reporting origin rather than independent confirmation.
The reported leadership change
Hu joined Tencent in early 2025 after serving as a principal researcher in Microsoft Research Asia's Visual Computing Group, according to 36Kr. At Hunyuan, he worked on vision foundation models before moving into a frontier-research group inside the LLM department, where he oversaw multimodal understanding and also participated in world-model work.
Tian joined the department in early July after working at OpenAI. The report says he will lead vision-language-model R&D following Hu's departure. It does not describe a public product launch, a completed team closure or a finalized new organization chart.
World models are a possible direction
36Kr reported that Yao has been reviewing teams and that Hu's former group may focus on frontier world-model research. That is a possible direction, not a confirmed Tencent roadmap.
World models generally learn representations of an environment and predict how it may change, linking perception with planning or action. The label covers multiple research approaches. The available reporting does not identify a specific architecture, training corpus, benchmark or release schedule for a Tencent system.
The report also argues that Hunyuan is concentrating talent and compute around foundation-model R&D. It says Tencent previously dissolved its AI Lab and moved core researchers into the LLM department, while hiring staff with infrastructure and post-training experience. Those organizational details come from the same 36Kr reporting and remain unattributed by Tencent.
What practitioners can conclude
The supported conclusion is narrow: Hunyuan's multimodal leadership is changing, Tian is expected to assume the VLM remit, and one research group may shift toward world models. It would be premature to infer that Tencent is abandoning multimodality or has committed to a particular world-model product.
For ML organizations, the case illustrates how leadership changes can signal resource prioritization without yet supplying technical evidence. The next verifiable signals would be a Tencent announcement, a model release, a paper, hiring changes or benchmarks showing where the group is actually investing.
Key Points
- 136Kr reports that Hu Han tendered his resignation from Tencent Hunyuan and that Tian Yonglong will take over vision-language-model R&D.
- 2Hu's former group may shift toward world-model research, but Tencent has not confirmed a roadmap, architecture or product.
- 3KrASIA adapted the 36Kr report, so the two retrieved hosts do not constitute independent reporting origins.
Scoring Rationale
The reported transition affects research leadership at a major Chinese model developer and may indicate resource movement toward world models. Its technical implications remain speculative, Tencent has not confirmed the report and both retrieved articles trace to one reporting origin.
Sources
Public references used for this report.
Practice with real Ad Tech data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Ad Tech problems
