Yao Shunyu calls for backup

"Could this be a highly significant year in the history of Tencent's AI?"

With the release of the new model Hy4, a wave of discussions has emerged. Prior to this, Tencent Hunyuan used to present its models in a down-to-earth and humble manner, but this time, the official release blog stated confidently: Hy4 has joined the ranks of the world's first-tier open-source models.

The confidence comes from Yao Shunyu. This young man born in 1998 was officially appointed as Tencent's Chief AI Scientist at the end of 2025. At that time, he took over Hy2, which still had insufficient capabilities, and led the launch of Hy3 preview, Hy3, Hy4 preview, and then Hy4. In just one year, Tencent has seemingly become the internet giant with the most rapid improvement in model capabilities.

From this, the right person is of utmost importance. On the same day of Hy4's release, Tencent Hunyuan, led by Yao Shunyu, also launched recruitment, including social recruitment and the Qingyun Project — according to the official recruitment website, the Qingyun Project covers more than 160 positions, while Hunyuan has over 60 positions for social recruitment.

The AI competition is a long marathon. Tencent undoubtedly has a deep understanding that success depends on people.

Tencent is catching up and ramping up its recruitment

The outside world has begun to feel that Tencent's AI is truly at the core of the competition table now.

The newly released Hy4 further expands the scale of pre-training and post-training. Compared with the Hy4 preview three months ago, Hy4 "opens its eyes" for the first time and introduces visual understanding capabilities. In terms of parameter scale, the total parameter scale of Hy4 doubles compared to Hy3 to 770B, and it supports computer operation for the first time.

Professional financial analysis is one of the key upgrades. Hunyuan invited 213 financial experts to conduct 579 real task crowdsourcing tests around scenarios such as fund research, securities analysis and quantitative strategies. The results show that the official version of Hy4 scored 3.04 in total, ranking first among the participating open-source models.

This time, the official directly put forward the keynote slogan "Born for real productivity". As the head of Hunyuan, Yao Shunyu did not hide his affirmation when reposting on social platforms: I am proud of the achievements we have made in the past year.

However, the Hunyuan team still remains modest, stating that although Hy4 is already one of the strongest open-source models at present, there is still a large gap compared with top closed-source models, and it will continue to steadily expand the scale of pre-training and post-training to strive to narrow the gap.

At this much-watched moment, Tencent Hunyuan has conveniently released a job recruitment notice:

For the Qingyun Project, it recruits campus hires and interns, with demand departments including the Basic Model Department, AI Infra Department, and AI Data Department. Its official recruitment website shows that there are 165 positions available for recruitment; social recruitment positions include the Basic Model Department, AI Infra Department, etc., and the official website shows that there are more than 60 positions available for recruitment. We noticed that the vast majority of social recruitment position information was updated from October to November 2026.

Expanding the team is nothing unusual. Over the past year and more, Tencent Hunyuan, which is in a catch-up state, has continued to recruit talents with high salaries, and core talents such as Tian Yonglong and Lin Xudong have joined one after another. The Information once reported that Tencent actively poached some core researchers from ByteDance with double salaries, and offered new doctoral graduates 50% higher than the industry average. In July this year, Yao Shunyu himself also recruited for Hunyuan AI4S on social platforms, saying "Hyai4S is hiring".

Now that Hy4 has caught up, the demand for talent density will not stop. But a change can be felt: since the second half of this year, the recruitment requirements of domestic basic model teams including Hunyuan have increased significantly, "it is harder to get in this year".

Yao Shunyu's past year

It has only been within this year that it has rapidly reached the level of the domestic first tier.

Yao Shunyu is considered the figure who reversed the situation. Before that, Tencent was the last domestic major manufacturer to launch its self-developed large model, which once made Tencent fall into the anxiety and embarrassment of not getting the AI ticket. At the shareholders' meeting in May this year, even Ma Huateng reviewed: "A year ago, we thought we had boarded the ship, but later found that the ship was leaking".

It was not until the end of 2025 that a decision full of transformative significance was made — Tencent officially announced the appointment of Yao Shunyu, born in 1998, as the Chief AI Scientist of the CEO/President's Office. The resume of this genius teenager quickly attracted the attention of the outside world:

A native of Hefei, Anhui Province, Yao Shunyu studied at Hefei No.1 High School in high school. In 2015, Yao Shunyu entered the Yao Class of Tsinghua University with a score of 704 in the college entrance examination, majoring in computer science. He studied under Wu Jiajun, a senior at Tsinghua University and a deep learning expert, and began to systematically get in touch with general artificial intelligence;

In 2019, Yao Shunyu went to Princeton University to pursue his doctorate. He originally focused on computer vision, but Yao Shunyu believed that natural language processing had more potential, so he switched to natural language processing and reinforcement learning, and got acquainted with his supervisor Karthik Narasimhan. At this time, the famous Transformer architecture had been born for two years, and natural language processing was a hot direction.

After graduating with a doctorate, Yao Shunyu joined OpenAI in 2024 and participated in the research and development of key projects such as Operator, DeepResearch and computer-using agents. When Tencent poached Yao Shunyu from OpenAI, he actually only had one year of work experience.

But Tencent was very certain and provided this AI talent with an unprecedented salary level and onboarding specification.

Yao Shunyu joined Tencent in December 2025. Three months later, the 10-year-old AI Lab was revoked, and the core R&D personnel were merged into Hunyuan. In July this year, the Large Language Model Department and the Multimodal Model Department were merged to establish the Basic Model Department, which is uniformly managed by Yao Shunyu. Now, in addition to managing the two major departments of the Basic Model Department and AI Infra Department, Yao Shunyu also serves as the head of the AI Data Department, fully covering the full link of the Hunyuan large model.

In less than a year, Yao Shunyu began to hand in his report card.

Hunyuan Hy3 released in July can be regarded as the first report card after Yao Shunyu joined Tencent. At that time, not only inside Tencent, but also the outside world's expectation for Yao Shunyu and Hunyuan Hy3 was only "getting out of the trough and stepping onto the track". The current Hy4 shows more meaning of starting to demonstrate its strength and even sincerity.

"It has been greatly improved compared with Hy3 and even Hy4 preview, and there is no suspense that it ranks in the top 3 of domestic models", an industry insider exclaimed after actual measurement. The outside world expects that if nothing unexpected happens, the total parameter scale of Hy5 will reach the tera-parameter level.

Full-scale competition

The AI competition is a long marathon.

For Tencent, facing the second half of AI where the market pattern is not yet determined and the business model is still evolving, after the basic model catches up, it should give full play to the scenario, its thickest hidden card — as an internet giant with a large number of To C and To B products, Tencent connects real users, specific scenario demands and business processes.

The most obvious undertaking is WorkBuddy, an AI intelligent office assistant launched by Tencent in March this year. It is reported that more than 50% of China's top 500 enterprises are now using WorkBuddy, and 60% of the top 100 enterprises have been covered. Inside Tencent, WorkBuddy is regarded as the next strategic product after QQ and WeChat.

"It is important to start early, but only by persisting for a long time can you have the opportunity to seize the new opportunities that keep emerging." Tang Daosheng, Senior Executive Vice President of Tencent Group, said recently when responding to the comment that Tencent's AI "moves too slowly".

After all, there is no permanent leader in the AI era.

We still remember that two years ago, all manufacturers were competing in parameter scale, training data and computing power investment, with the competition focus on reasoning, code, long text, image and video generation. Then in 2025, models such as DeepSeek-R1 and OpenAI o-series made complex reasoning the focus of competition, and the industry began to pay attention to call price, response speed and actual effect.

Today, AI competition is shifting from simply pursuing model capabilities to all-round competition among models, agents, infrastructure, commercialization and security governance. The most intuitive manifestation is that AI investment is becoming increasingly generous.

Just looking at domestic giants, the market rumor says that ByteDance's 2026 AI investment budget is close to 1.6 trillion yuan, Alibaba announced that it will invest 380 billion yuan in AI and cloud computing within three years and increase capital expenditure, and Tencent's investment in AI infrastructure in a single quarter also exceeds 100 billion yuan. DeepSeek spent about 11 billion yuan on AI infrastructure-related expenditures in the first 7 months of this year, a 10-fold surge compared with 2025.

After chasing the scale of investment, people will inevitably begin to examine the return. For the young people in charge of AI in major manufacturers, the water is wide and the fish are big, the wind is strong and the rain is heavy. This competition that affects the fate of giants has just begun.

This article is from the WeChat official account "Investment Circle" (ID: pedaily2012), written by Feng Yuchen, and authorized for release by 36Kr.