Tencent Merges Its Hunyuan Teams Into One Foundation Model Department

On July 23, 2026, Tencent announced that its Hunyuan large language model team and its Hunyuan multimodal model team will be combined into a single foundation model department. Both groups had been run as separate units since April 2025, when Tencent first split them to let each side focus on its own area. The new department will report to chief AI scientist Yao Shunyu, and the goal is to push research and engineering toward unified, all-modality models. The reorganization comes after a long internal rebuild. Since the end of January 2026, Hunyuan has been reworking its pretraining stack, reinforcement learning pipeline, data system, and evaluation framework from the ground up. In July, the team released Hy3, a new language model that Tencent says is smarter than other models of similar size and on par with flagship systems two to five times larger. Putting text, image, and audio work under one leader is meant to turn that progress into shared infrastructure. Tencent's move reflects a wider shift in how Chinese AI labs are structured. As training compute and data costs keep rising, more companies are pulling their model research back into a single organization instead of letting separate groups compete for the same GPUs. A unified team can share pretraining code, evaluation tools, and safety work, and it makes it easier to ship a single model that handles text, vision, and audio together. For the rest of the industry, the Hunyuan merger is a signal that the next phase of model competition is no longer just about bigger benchmarks. It is about whether a lab can keep research, infrastructure, and product tightly aligned. Chinese tech companies that run very large consumer apps, from messaging to video, now have a strong reason to build one foundation model team rather than several, because every product team wants the same underlying model to plug into.