Tencent's AI Lab, founded in 2016, spent a decade publishing numerous papers but was abruptly disbanded in March 2026. Employees have been transferred en masse to the Hunyuan large-model team, working alongside a 28-year-old chief AI scientist to finalize the launch of Hunyuan 3.0 by the end of April.
This move itself is telling: an academically oriented AI Lab, at this juncture, can no longer easily sustain its independent existence.
Who Is Yao Shunyu
The key figure taking over the Hunyuan team is Yao Shunyu, recruited by Tencent in December 2025 as chief AI scientist. Before joining Tencent, he worked at OpenAI.
He is best known for designing the ReAct framework (an agent architecture that interleaves reasoning and action) and Tree of Thoughts (a method that enables large models to think like a decision tree). Both frameworks have high citation counts in the LLM agent field and are considered genuinely influential work, not just publications for the sake of publishing.
Yao earned his undergraduate degree from Tsinghua's Yao Class, the elite computer science program founded by Turing Award winner Andrew Yao. Securing a position at this level at age 28 is uncommon, even in the 2026 AI talent market.
He has a publicly stated philosophy: drive products with scenarios, not with benchmark scores. This clearly distinguishes him from most domestic AI teams currently focused on competing on benchmarks. Whether he can deliver, of course, depends on the results of Hunyuan 3.0.
Hunyuan 3.0: Parameter Scale Is Not the Focus
Hunyuan 3.0 has approximately 30 billion parameters and is already in internal testing, with a public release planned for this month.
Compared to parameter scale, this release emphasizes two directions:
- Enhanced reasoning capability: Targeting the reasoning model track, Hunyuan aims to upgrade from answering questions to actual thinking.
- Agent usability: Yao prioritizes agent practicality as the primary direction; Hunyuan 3.0 must function in real workflows, not just look good in demos.
If both directions are executed well, Hunyuan 3.0 would be positioned not just as another domestic large model, but as a tool with practical deployment value in the Chinese enterprise market.
This positioning directly competes with ByteDance's Doubao and Alibaba's Tongyi Qianwen. As the domestic reasoning model race intensifies, delivering convincing agent capabilities is the key metric.
WeChat AI Agent: Tencent's Real Trump Card
Hunyuan 3.0 is certainly not Tencent's most important card in AI. WeChat is.
Tencent is developing an independent AI agent entry point for WeChat. The goal: leverage the existing ecosystem of WeChat mini-programs, WeChat Pay, and official accounts to let users complete tasks that previously required jumping between multiple apps—finding a restaurant, booking a table, paying, and sharing—all within a single dialog.
Timeline: a trial run in the domestic WeChat (Weixin) is expected in mid-2026, with full rollout in the third quarter.
This logic aligns with Tencent's previously launched QClaw (a WeChat ecosystem AI agent tool), but the scale is incomparable. WeChat has over 1.4 billion monthly active users, a channel advantage no standalone app can replicate.
ByteDance's Doubao still relies on paid traffic for users, but WeChat's users are naturally accumulated. If Tencent builds a good enough agent, this channel advantage becomes a game-changer.
Yuanbao's 50 Million DAU: Numbers Have Caveats
Tencent announced in February that its AI assistant Yuanbao had surpassed 50 million daily active users.
This figure sounds impressive, but the context is that Tencent spent 1 billion RMB on red-packet campaigns during the Spring Festival, driving user numbers through massive traffic investment. Post-holiday data showed a significant decline.
This does not mean Yuanbao lacks a foundation, but it indicates that user retention and DAU quality have not yet stabilized. Whether Tencent can convert WeChat's natural traffic into genuine sticky users for Yuanbao is what truly needs to be proven.
The Essence of This Restructuring
Tencent's AI Lab has been evaluated internally as having done solid foundational research but consistently weak in product conversion. This relates to Tencent's overall traffic-driven business DNA and the Lab's incentive structure—publishing papers and attending top conferences was less rewarding than building products in a more engineering-focused environment.
By fully merging the Lab into the Hunyuan team, Tencent is making a clear statement: research must serve products; it cannot live in its own world anymore.
With Yao Shunyu, who has both theoretical depth and agent deployment expertise, leading the charge, Tencent's direction seems sound.
But on the execution front, ByteDance's Doubao, Alibaba's Tongyi, and Baidu's ERNIE are not waiting. By 2026, the domestic large-model competition has entered a zero-sum phase. Whether Hunyuan 3.0 can deliver a convincing result by the end of April will be the decisive battle for what comes next.
Sources: CocoLoop, Tencent to Launch Hunyuan 3.0 in April, Build WeChat AI Agent (Caixin Global); Tencent Folds AI Lab Into Hunyuan Team in Major AI Overhaul (Caixin Global); DeepSeek V4 And Tencents New Hunyuan Model To Launch In April (Dataconomy)