Tencent Hunyuan Organizational Restructuring: Old Guard Departs After Yao Shunyu Takes the Helm

In the third year of the large model rush, the internal organizational structures of the major tech companies have finally undergone their first wave of substantive, deep restructuring.

Recently, news about organizational restructuring and partial personnel changes in Tencent's Hunyuan team has sparked discussion on professional social networks and within the AI community. Many employees' surprise and dismay are written all over their faces. This sense of surprise is not hard to understand: after all, not long ago, at the HY4 Preview release milestone, the team had just gone through a round of high-intensity sprinting and a brief vacation; and moreover, because Yao Shunyu, a former core OpenAI researcher regarded by the outside world as Tencent AI's "new commander," had just arrived to lead foundation model R&D. To outsiders, this should have been a moment for sharpening weapons and expanding on a large scale—so why has it instead brought personnel contraction?

One factual boundary must first be clarified: Tencent officially has never characterized the changes at Hunyuan as "layoffs" or "large-scale personnel optimization." At Tencent's shareholders meeting in May 2026, President Liu Chiping explicitly responded that "there is definitely no major layoff plan; Tencent is not quite the same as Silicon Valley companies." What is happening at Hunyuan is not a top-down layoff campaign, but an organizational restructuring after Yao Shunyu took office, characterized by department mergers, compressed reporting lines, and partial team personnel replacements. The difference between the two is not a word game, but the key to understanding the direction of Tencent's AI strategy.

If one looks only at the comings and goings of a certain team, it is easy to misread them as a setback for a single business. But from the macro perspective of Tencent's overall AI strategy and the cycle of China's large model industry, this "succession of the new for the old" is not only inevitable, but can even be said to be the path Tencent must take to redeem itself in the deep waters of AI.

This is not merely a simple personnel change, but a comprehensive switch by Tencent's large models from "extensive land-grabbing with a human-wave tactic" to "saturation attacks by a core elite force."

The reckoning after the frenzy: squeezing the "water" out of large model teams

Looking back at 2023 and 2024, that was the golden age of the entire Chinese large model track. In order to quickly keep up with the technological wave triggered by ChatGPT, the major tech companies launched a cost-no-object arms race.

At that stage, the expansion of large model teams was extensive and brute-force. Tens of thousands of algorithm engineers, data annotators, and evaluators were recruited at high salaries. The industry consensus at the time was that "brute force creates miracles"; in order to quickly close the generational gap with overseas giants, the major companies needed to build, in the most efficient way, a complete pipeline from data cleaning, manual labeling, and basic fine-tuning to model evaluation. During this process, the Hunyuan team expanded rapidly, absorbing a large number of junior engineers and middle managers.

However, by 2026, the situation had fundamentally reversed. The underlying technical paradigm of large models had become relatively stable, and the marginal effect of the Scaling Law began to diminish. When the parameter scale of models reached the trillion level, simply stacking computing power and increasing data volume could no longer bring stunning qualitative changes. At the same time, the extreme inflation of computing costs forced the CFOs of the major companies to turn their attention to labor costs.

Even more fatal is that, with the improvement of large models' own capabilities and the maturation of internal toolchains, those "dirty and tiring jobs" that previously required a large amount of manual work—such as basic data cleaning, routine fine-tuning tests, and adaptation to edge scenarios—are being replaced by automation and more advanced models.

As the frenzy recedes, the major companies begin to do the math. As the large model business moves from "proving we can build it" into the second half of "proving we can make money," efficiency pressure is inevitably transmitted to the most cash-burning departments. For the major companies, foundation model R&D has become an oligopoly game; continuing to maintain a huge regular army not only yields low marginal returns but also causes severe organizational bloat. Cutting redundant positions in marginal business lines and tilting resources toward the most core assault teams is the only solution for maintaining competitiveness.

The arrival of the new commander and the rejection reaction of technology

Behind this organizational contraction closely follows a drastic reconstruction of the technical roadmap.

Yao Shunyu's joining is an extremely rare big move by Tencent in its AI personnel lineup. Born in 1998, a graduate of Tsinghua's Yao Class and a Princeton PhD, he joined OpenAI in 2024 to participate in agent product R&D, and in December 2025 joined Tencent as Chief AI Scientist of the "CEO/President's Office," reporting directly to Liu Chiping, while also serving as head of the AI Infra Department and the Large Language Model Department. In July 2026, Tencent merged the Large Language Model Department and the Multimodal Model Department into the Foundation Model Department, with Yao Shunyu managing it uniformly and coordinating the language, multimodal, reinforcement learning, and Agent directions, making him the de facto "number one" in Hunyuan's technical R&D system. To the outside world, this is the "reinforcement" Tencent brought in to turn the tide in the next-generation models (HY4 and subsequent versions). The arrival of a top scientist inevitably brings the most cutting-edge technical vision, training methodologies, and a rethinking of the underlying architecture.

But beneath the halo of the scientist lies the cold metabolism of the organization.

The establishment of a new technical roadmap often means the end of an old R&D logic. If Yao Shunyu is to reconstruct Hunyuan's underlying technology stack and relentlessly tackle native multimodality, complex logical reasoning (CoT), and more efficient reinforcement learning (RL) frameworks, then the knowledge structure of the original team will face severe challenges.

Teams that in the past were only responsible for fine-tuning specific business lines, or focused on maintaining legacy architectures, find it hard to locate their place in the new blueprint. A new leader's ascent is often accompanied by the exit of the old guard. This is not personal workplace infighting, but an unavoidable "rejection reaction" in technological evolution. If Tencent wants to build a "special forces" unit that can resonate with OpenAI and DeepMind, it must streamline a bloated chain of command and concentrate resources in the hands of a few top-tier "generals," rather than continuing to feed a conventional army that relies on sheer headcount to pile up leaderboard scores.

From "chasing benchmark scores" to "chasing deployment": Tencent's strategic pragmatism

Reviewing the evolution of the Hunyuan team's positioning within Tencent makes the internal logic of this round of adjustment clearer.

In the first half of the hundred-model war, Hunyuan's core task was to "ensure it didn't fall behind." But in the second half, Tencent's consideration of AI has shifted comprehensively toward application deployment and ROI (return on investment).

Recently, Tencent's moves in AI have been extremely pragmatic. On the B2B side, WorkBuddy, leveraging the ecosystem advantages and productization capabilities of WeCom, has established a leading position in the office agent track. According to QuestMobile data, in September 2026 WorkBuddy's combined monthly active users across platforms reached 32.353 million, ranking first in the typical office Agent category, with quarter-over-quarter growth of 647.1%. On the consumer side, although Yuanbao faces strong suppression from Doubao—in September 2026 Doubao had 419 million MAU versus Yuanbao's 37.33 million, and the two are no longer in the same league—Tencent has not blindly followed the trend of burning money to buy traffic, but instead embedded AI capabilities more deeply into national-level apps with massive daily active users, such as WeChat, QQ, and Tencent Meeting.

This shift in strategic focus directly determines the Hunyuan team's talent needs.

In the past, large-model R&D was "technology-driven," with everyone competing on benchmark scores and parameters; now, large-model R&D is "scenario-driven," competing on who can solve specific business pain points at lower inference cost. Tencent no longer needs so many people who can only write general-purpose algorithms; it urgently needs composite talents who understand both large models and business scenarios, and can seamlessly stitch AI into the WeCom and WeChat ecosystems. Those employees unable to adapt to this shift from "technology-oriented" to "business-oriented" naturally became the cost of this round of organizational adjustment.

This wave of adjustment at Tencent Hunyuan is not only a correction of its own strategy, but also a microcosm of the entire Chinese large-model industry moving toward maturity.

Not just Tencent—Alibaba, ByteDance, Baidu, and other major companies, after the aggressive expansion of the past two years, are all now to varying degrees "shedding fat and building muscle" in their large-model teams. In the first half of 2026, all four giants completed the move of "merging scattered model teams into one core department": ByteDance's Seed continued integrating, Alibaba established the Token Foundry business unit, Tencent formed the Foundation Model Department, and Baidu advanced BMC integration. The rules of the game across the industry have changed: the phase of blindly pursuing model parameter scale and team headcount is over, and the focus of competition going forward is extreme inference cost control, deep data moats, and killer application scenarios.

For the employees who left Hunyuan in this round of adjustment, the personal sense of loss is real. But from an objective view of industrial development, these engineers—carrying big-tech hands-on experience and having been tempered by thousand-card/ten-thousand-card cluster training—flowing into AI applications across thousands of industries has instead accelerated the spread of technology. They are no longer just "test-takers" deducing parameters in the cloud, but will become the sparks that truly monetize AI technology in the business world.

For Tencent, this pain is something it must endure. Large-model R&D is no longer a game that can go live after writing a few lines of front-end code; it is a marathon about underlying compute, top talent, and ecosystem efficiency. At this critical juncture when Yao Shunyu is in place and the old is giving way to the new, only by pulling the fist back and using the most elite team to relentlessly tackle the most core technology can Tencent ensure it remains at the table in the coming AI decisive battle.