Tencent's Hu Han Leaves to Found Startup; World Models May Become Focus
Hu Han, who led Tencent's Hunyuan multimodal understanding efforts, has resigned to start his own company. Previously a chief researcher at Microsoft Research Asia's Visual Computing Group, Hu joined Tencent in early 2025 to focus on visual large models. He later moved to the 'Frontier' research group within the large language model department, reporting to Yao Shunyu, where he concentrated on multimodal understanding and was also involved in developing world models. Following Hu's departure, the research group he led is reportedly shifting its focus to cutting-edge world model research under Yao Shunyu's direction. This organizational shift is part of a broader strategy by Yao Shunyu, head of Tencent's large language model department, to consolidate resources and accelerate the development of foundational models. Since mid-2026, Tencent has been restructuring its AI teams, including integrating Tencent AI Lab's core researchers into the large language model department and increasing recruitment efforts for AI talent. This strategic realignment aims to bolster Tencent's competitive position in the rapidly evolving AI landscape, with the recent release of the Hy3 large model marking a significant step. The decision to de-emphasize multimodal understanding research, which is nearing maturity with limited new gains, is attributed to a strategic re-evaluation of resource allocation and potential returns. While multimodal understanding has historically been a strong area for Tencent, its direct commercialization potential is seen as less clear compared to generative AI capabilities. With limited computing resources compared to competitors like Alibaba and ByteDance, Tencent is prioritizing areas with higher perceived future value and clearer pathways to monetization, such as advanced reasoning and agentic capabilities.
AI development at Tencent appears to be undergoing a strategic pivot, prioritizing foundational model advancements and potentially world models over mature multimodal understanding research. This shift reflects a common industry trend where companies re-evaluate R&D investments to align with emerging technological frontiers and clearer commercialization paths, especially given resource constraints. The emphasis on consolidating talent and resources into core large language model development suggests a drive to achieve parity or leadership in a highly competitive AI ecosystem. The internal urgency noted by researchers, coupled with strict directives on data quality, indicates a management focus on rapid progress and execution. Evaluating future R&D investments against potential market impact and technological readiness will be crucial for Tencent as it navigates the next decade of AI evolution, where foundational capabilities and novel architectures like world models may redefine the competitive landscape.
AI-generated to prompt reflection — not editorial opinion, not advice, not a statement of fact. How this works.