Tencent Merges AI Teams Into New Unit Led by Chief Scientist
Tencent Holdings has consolidated its multimodal and large language model teams into a new Basic Model Department, placing all foundational artificial intelligence research under Chief Scientist Yao Shunyu in a sweeping restructuring that marks the end of the company’s decentralized “horse-racing” approach to AI development.
The reorganization, announced on July 23 and reported by Caixin Global, merges the Hunyuan Multimodal Model Department and the Large Language Model Department into a single unit, dismantling the dual-track system that had been in place since April 2025. The new department reports directly to Lu Shan, President of Tencent’s Technology and Engineering Group (TEG).
A Strategic Shift Under Yao Shunyu
The restructuring represents the culmination of a year-long consolidation drive led by Yao Shunyu, a Tsinghua University graduate and former OpenAI researcher who joined Tencent in September 2025. Appointed Chief AI Scientist in December 2025, Yao has systematically centralized Tencent’s previously fragmented AI operations.
In March 2026, Tencent dissolved its AI Lab and reassigned staff to the Hunyuan teams. The latest merger brings language, vision, audio, video, and 3D generation capabilities under a single technical roadmap for the first time.
According to an internal Tencent notice cited by Caixin, the new Basic Model Department is focused on “frontier exploration, model architecture, data strategies, reinforcement learning, and agent capabilities.” The unit is tasked with shepherding Tencent’s Hunyuan series of AI models, which power everything from chatbots to image and video generation tools.
End of the Horse-Racing Era
Tencent’s approach to AI had historically been characterized by internal competition, with multiple teams pursuing parallel research tracks. This “horse-racing” strategy allowed Tencent to explore diverse approaches but created duplication and coordination challenges.
The dual-track system established in April 2025 split the Hunyuan team into separate LLM and multimodal departments, each pursuing independent research agendas. After just 15 months, Tencent has opted to re-merge these teams under unified leadership — a move that analysts say reflects both the maturing of AI technology and the intensifying competitive pressure facing the company.
“The key to the second half of AI is to build a long-term organization oriented toward AGI,” Yao has said, outlining a vision that prioritizes sustained research investment over short-term benchmark chasing.
Competitive Landscape
The restructuring comes at a critical juncture for Tencent’s AI ambitions. The company faces fierce competition from Chinese tech rivals including ByteDance, which leads in video-generation AI, and Moonshot AI, whose Kimi K3 large language model has gained significant traction. Moonshot AI’s success was underscored on July 16 when Kimi K3 demand forced the startup to temporarily pause new sign-ups.
Tencent has responded aggressively on the product front. In July 2026 alone, the company released Hunyuan Hy3 — a Mixture-of-Experts model with 295 billion total parameters that delivers a 40% improvement in inference efficiency and reduces hallucination rates from 12.5% to 5.4% — launched the Miora image-generation agent, and merged its QClaw product into its WorkBuddy platform.
The Hunyuan Hy3 release on July 6 drove a 68-fold surge in usage and boosted Tencent’s stock by 11%, demonstrating the market’s appetite for the company’s AI progress.
Implications and Outlook
The consolidation of AI teams under Yao signals a shift from research exploration to product-focused execution at Tencent. By unifying model development across modalities, Tencent aims to accelerate development cycles and create more cohesive AI products that can compete with offerings from ByteDance, Moonshot AI, and global players.
Yao has also pursued an open-source strategy, releasing HyWorld 2.0 in April 2026 to more than 3 million downloads, and has recruited talent from ByteDance, Alibaba, and China’s AI startup ecosystem to strengthen Tencent’s capabilities.
Outstanding questions remain about the future of former department heads displaced by the restructuring. Bo Liefeng, who led the Multimodal Model Department, and Hu Han, the former multimodal understanding lead, are both believed to have departed, though their next moves have not been confirmed.
As China’s AI race intensifies, Tencent’s bet on centralized leadership and unified research under Yao Shunyu represents a high-stakes gambit. If successful, the restructuring could position the company to compete more effectively across the full spectrum of AI capabilities. If not, the loss of internal competition could slow the innovation that Tencent’s horse-racing model once fostered.