Tencent Merges Multimodal LLM Teams Into New AI Unit

Tencent Holdings Ltd. has consolidated its multimodal and large-language-model teams into a single Foundation Model Department, placing the unified unit under the leadership of Chief AI Scientist Yao Shunyu. The restructuring, announced internally on July 23, marks the end of a dual-team parallel development model and signals Tencent’s intensified push to compete in the rapidly evolving multimodal AI landscape, according to Caixin Global.

Context: A Strategic Consolidation

The new Foundation Model Department merges Tencent’s Hunyuan Large Language Model (LLM) Department and its Multimodal Model Department, which had operated as separate entities since April 2025. The company stated the reorganization aims to improve model research and development efficiency while exploring what it calls the “intelligence ceiling” of full-modality models.

Yao Shunyu, who joined Tencent from OpenAI in September 2025 and was appointed Chief AI Scientist in December of that year, now holds full authority over all underlying model R&D spanning text, image, video, speech, and 3D generation. As National Business Daily (NBD) reported, his scope of responsibility has “significantly expanded” to encompass Tencent’s entire full-modality research pipeline.

Key Developments: Accelerated AI Investment

The restructuring builds on a series of moves that have reshaped Tencent’s AI strategy over the past year. In March 2026, Tencent dissolved its long-standing AI Lab, reassigning staff to the Hunyuan team under Yao’s direction. The company has also aggressively recruited top talent: former OpenAI researcher Tian Yonglong, a former colleague of Yao, joined Tencent’s LLM department on July 8, while “post-95s” scientist Pang Tianyu had joined the multimodal team in January.

Just weeks before the restructuring, Tencent released Hy3, a Mixture-of-Experts (MoE) architecture LLM with 295 billion total parameters (21 billion active). According to Zhidx, Hy3 demonstrated a 20-30% improvement in agent and code capabilities over its April preview, delivering performance comparable to models two to five times its scale. The model saw remarkable adoption: first-week call volume grew 68-fold over its predecessor Hy2, and 60% of WorkBuddy self-model users chose Hy3.

Tencent’s Hunyuan ecosystem now includes a complete model matrix. The open-source Hunyuan 3D World Model 2.0, released in April 2026 and supporting text, image, and video inputs, has garnered over 3 million community downloads. Meanwhile, Hunyuan Image 3.0 ranks first among domestic open-source models on the LMArena benchmark.

Analysis: Competitive Pressure Drives Consolidation

The restructuring comes amid intensifying competition in China’s AI sector. Rival Moonshot AI launched its Kimi K3 model on July 16, a model praised by Elon Musk that drove user demand so high it pushed computing clusters close to capacity within 48 hours. Tencent’s stock fell 4.6% on July 17 following the Kimi K3 launch, likely reflecting market concerns about competitive pressure.

ByteDance has captured a leading position in video generation models following OpenAI’s suspension of Sora, while Tencent has also been investing externally — deploying 1.4 billion yuan ($210 million) for a 1.1% stake in Kuaishou Technology’s Kling AI.

Industry observers view the restructuring as a necessary response to these dynamics. By centralizing authority under Yao Shunyu, Tencent eliminates intra-company competition between its LLM and multimodal teams, streamlines decision-making, and creates a unified path toward its stated goal of artificial general intelligence (AGI).

What’s Next: Full-Modality Ambitions

Tencent’s unique advantage in the AI race may lie in its gaming business, which provides rich 3D assets, rendering expertise, and natural use cases for 3D world models. The company’s continued talent acquisition from OpenAI — with both Yao and Tian Yonglong bringing frontier research experience — suggests Tencent is building a world-class research capability.

As Yao Shunyu consolidates control over Tencent’s entire model development pipeline, the key questions ahead include what specific milestones will measure the success of this unified structure, how Tencent will coordinate internal development with external investments like Kling AI, and whether the company can translate its full-modality ambitions into products that close the gap with rivals ByteDance and Moonshot AI.

The creation of the Foundation Model Department represents Tencent’s most decisive organizational commitment to AI yet — placing a single leader and a unified team at the center of its efforts to define the next generation of intelligent systems.