Tencent's latest large model, detailed explanation of Tencent's full picture
Introduction:Tencent $TENCENT (00700.HK)$ Has the newly released HunYuan Hy3 preview language model already demonstrated Yao Shunyu's capabilities?
Lishi, Yuanbao/Author Lishi Business Review/Publisher
Since former OpenAI star researcher Yao Shunyu joined Tencent to lead the company’s large-scale model development, the first result of his efforts has been highly anticipated by outsiders.
Just recently, the HunYuan Hy3 preview language model, primarily developed under Yao Shunyu’s leadership, was officially released and announced as open source, with early adoption in Tencent's proprietary products such as Yuanbao, IMA, WorkBuddy, and CodeBuddy.
How does the Hy3 preview language model perform? Is it an improvement over Tencent's previous large models? To answer this question, the 'Lishi Business Review' specifically tested the Yuanbao App, which applies the Hy3 preview language model, by asking various questions related to Tencent. The responses from the Yuanbao App were fairly standard; although still lacking some “human touch,” they demonstrated strong accuracy, systematic thinking, and logical coherence, with some insights providing valuable takeaways. We have shared these Q&A sessions below so that readers can get a general sense of the Hy3 preview language model while also gaining a macro understanding of Tencent’s overall business operations.
Lishi Business Review:Please provide an overview of Tencent's business operations.
Yuanbao App:Tencent’s business operations are extremely broad. Below are its main business segments:
I. Value-added Services
1. Online Games
Tencent is the world's largest game publisher, with numerous popular mobile games such as Honor of Kings and Peacekeeper Elite, along with a steady stream of new game releases.
2. Social Networking Services
This includes live streaming services on Video Accounts and paid video membership services. For instance, Tencent Video offers a wide range of movies, TV series, and variety shows to meet users' entertainment needs.
II. Online Advertising
1. Media Advertising
Utilizing media platforms under Tencent, such as Tencent News, to provide advertisers with ad placement services.
2. Social and Other Advertising Businesses
Based on social platforms like WeChat and QQ, advertisements are delivered to hundreds of millions of Chinese consumers through precise user targeting technologies.
III. Fintech and Enterprise Services
1. Commercial Payments
WeChat Pay is an essential part of Tencent's fintech offerings, widely used in various online and offline consumption scenarios, providing users with convenient payment options.
2. Fintech
Offers financial services such as wealth management and insurance to meet the diverse financial needs of users.
3. Cloud Services
Tencent Cloud provides enterprises with a range of cloud services including computing, storage, and databases, supporting companies in their digital transformation.
IV. Other Businesses
1. Investment Business
Tencent is an active strategic investor, holding stakes in leading internet companies such as PDD Holdings, Kuaishou, and Xiaohongshu.
2. Film and TV production and distribution
Producing and distributing movies and TV programs for third parties, such as participating in the investment and production of some popular film and TV dramas.
3. Content licensing
Licensing high-quality content to other platforms or enterprises for use, realizing the diverse value of the content.
4. Merchandise sales
Involves some merchandise sales activities related to Tencent's business.
Summary:Tencent enriches the lives of internet users through technology, connecting users via its social networking and communication platforms WeChat and QQ to a diversified range of digital content, including games, videos, music, and literature. At the same time, Tencent provides distinctive products in areas such as mobile payments, cybersecurity, cloud services, and artificial intelligence, promoting the business growth of its partners. It also heavily invests in talent and drives technological innovation, developing alongside the internet industry.
Lishi Business Review:How is the market performance of Tencent's various businesses?
Yuanbao App:Based on Tencent's 2025 earnings report and market data, all business segments of Tencent maintained steady growth in 2025, with total revenue reaching 751.77 billion yuan. Below is a detailed analysis of the market performance of each business segment:
1. Value-added services (gaming and social networks)
This is Tencent's largest revenue source, generating 369.3 billion yuan in 2025, representing a 16% year-over-year increase, accounting for 49.12% of total revenue.
Domestic market gaming:Revenue reached 164.2 billion yuan, marking an 18% year-over-year increase. Growth was primarily driven by the stable performance of perennial hits such as Honor of Kings and Peacekeeper Elite, along with a strong contribution from the new game Delta Force.
International market gaming:Revenue amounted to 77.4 billion yuan, reflecting a significant 33% year-over-year increase (32% at constant exchange rates). The growth was fueled by increased revenue from Supercell's games and PUBG MOBILE, as well as incremental income from the game Echo of Retribution.
Social networks:Revenue reached 127.7 billion yuan, showing a 5% year-over-year increase. This was mainly driven by revenue from live streaming services on Channels, paid music memberships, and sales of virtual items in mobile games.
2. FinTech and Enterprise Services (ToB business)
This segment is Tencent's second-largest revenue pillar, with revenue reaching 229.4 billion yuan in 2025, an 8% year-over-year increase, accounting for 30.52% of total revenue.
Financial Technology:Revenue achieved high single-digit percentage growth, primarily driven by increases in wealth management services, consumer loan services, and commercial payment activities.
Enterprise Services:Revenue grew nearly 20% year-over-year, reaching a new high. Growth was mainly driven by increased demand for cloud services (including AI-related services) and higher technology service fees due to the rise in WeChat Shop transaction volume. Additionally, Tencent Cloud achieved full-year profitability in 2025.
3. Marketing Services (Online Advertising)
This segment performed exceptionally well, with revenue reaching 145 billion yuan in 2025, a 19% year-over-year increase, accounting for 19.28% of total revenue.
Growth Drivers:The growth was mainly due to increases in both ad pricing and ad impressions. AI technology improved ad targeting precision, closed-loop ads (such as mini-programs and WeChat Shops) gained a larger share, and higher user engagement on Channels and WeChat Search contributed to the growth of the advertising business.
Technology enablement:Tencent upgraded its advertising foundational model and launched the intelligent ad placement product suite 'Tencent Advertising AIM+', achieving full automation from creative production to placement.
4. Other Business
Other operating income reached 8.077 billion yuan, accounting for approximately 1.07%, mainly including investment, film and television production and distribution, and content licensing businesses.
Profitability:Tencent's overall gross profit reached 422.6 billion yuan in 2025, up 21% year-on-year, with the gross margin rising to 56%. This was primarily driven by increased contributions from high-margin businesses such as self-developed games and video account advertising, as well as improved cost efficiency in cloud services and financial technology.
Stock price and institutional expectations:As of April 2026, Tencent (00700.HK) shares fluctuated between HKD 490-500. Institutions generally expressed optimism about its prospects, with approximately 95% of them giving a 'buy' or 'overweight' rating, forecasting that its revenue in the first quarter of 2026 would continue to maintain double-digit growth.
Lishi Business Review asks:Why has Tencent been able to achieve success across so many businesses and products?
Yuanbao App:The core reason why Tencent has managed to take the lead in fields as diverse as social media, gaming, finance, and cloud services lies in its creation of a 'traffic ecosystem + organizational mechanism + technological foundation' compound growth flywheel. Simply put, it is not just making products but running an expansive digital ecosystem.
Below are four key pillars supporting its diversified success:
1. Core Traffic and Ecosystem Synergy (Foundation)
Tencent owns two 'national-level' entry points, WeChat and QQ, which are not only social tools but also super incubators for all its businesses.
Connect Everything:WeChat has built a 'connect everything' system through features such as Official Accounts, Mini Programs, WeChat Pay, and Channels. New businesses (such as Channels and Mini Program e-commerce) can be seamlessly integrated into this ecosystem, enjoying a massive user base while significantly reducing customer acquisition costs.
Eco-loop:In the content sector, Tencent has connected the IP chain of Yuewen (literature), Tencent Video (film and television), and Tencent Games (interactive). A novel IP can be adapted into anime, TV series, and then developed into games, achieving a synergistic effect of 'multi-use of one fish.'
II. A Unique Organizational and Talent Mechanism (the Engine)
Tencent has a unique 'soil' mechanism internally that continuously produces blockbuster products.
Internal racing mechanism:This is Tencent's secret weapon for product innovation. When facing significant new opportunities, Tencent allows multiple teams to simultaneously develop similar products (the most classic case being WeChat, which outperformed other mobile communication products under development at the same time). This mechanism ensures that the best product emerges while avoiding decision-making paralysis in large companies.
Entrepreneurial Spirit:Tencent is able to retain top talent because it allows employees with an entrepreneurial spirit to start ventures within the company. Through sufficient empowerment and profit sharing, employees feel as if they are working for their own careers, which has spurred the creation of star products like WeChat and Honor of Kings.
3. Capital Linkages and Open Strategy (External Brain)
Tencent realized it couldn't excel in all细分 fields on its own, so it shifted its strategy from 'doing everything' to building an 'open ecosystem'.
The 'Half-Life' Theory:Ma Huateng proposed handing over 'half of Tencent's life' to partners. By investing (such as in JD.com, Meituan, and PDD Holdings) rather than fully acquiring stakes, Tencent has built a vast circle of allies within the 'Tencent ecosystem.' This not only brought substantial financial returns but also consolidated its indirect influence in areas like e-commerce and local life through payment and traffic empowerment.
4. Technology Platform and Adaptability (Fuel)
Technology Accumulation:Through technical departments such as TEG (Technology Engineering Group), Tencent has established a robust technology platform (such as cloud services, AI large models, and security capabilities). These underlying technologies can be reused across different business lines such as gaming, finance, and advertising, improving overall efficiency.
Strategic Iteration:Tencent has demonstrated a strong ability to self-evolve. From PC internet to mobile internet (WeChat) and then to industrial internet (the 930 transformation), it has always been quick to catch shifts in technology cycles and make early investments (such as going all-in on AI), avoiding being left behind by the times.
In summary, Tencent's success is not the victory of a single product but the result of the combined effect of 'social traffic x capital ecosystem x horse-race mechanism'.
"Lishi Business Review":What kind of management and cultural system has enabled Tencent to develop the aforementioned capabilities?
Yuanbao App:Tencent’s ability to succeed in so many businesses lies in its establishment of a management and cultural system characterized by 'extreme decentralization and high sharing.' Simply put, Tencent has divided itself into a group of 'internal ventures,' while everyone shares the same infrastructure for 'water, electricity, and coal.' This ensures both flexibility and the retention of resource barriers typical of large companies.
This system mainly comprises the following four core pillars:
1. Organizational Structure: Market-oriented Ecosystem Organization
Tencent adopts a 'federal' structure, making it appear like an alliance composed of countless small teams:
Business Groups (BG) enjoy high autonomy:Each business unit (such as WeChat, gaming, content) has independent budgets, technology, and decision-making power. The group primarily acts as a coordinator rather than a commander, enabling each business to respond quickly to the market like a venture.
Centralized Technology Sharing:Although business operations are decentralized, the underlying technology (such as the TEG Technology Engineering Group) remains unified. Each business team does not need to reinvent the wheel and can reuse company-level traffic platforms, technology platforms, and functional platforms, significantly reducing innovation costs.
Dynamic Adjustment Mechanism:Tencent has undergone several major organizational restructuring efforts throughout its history (such as the '930 Transformation' in 2018), flexibly reorganizing business groups based on market conditions (such as the transition from consumer internet to industrial internet), ensuring that the organizational structure remains adaptive.
II. Innovation Mechanism: Internal Competition and Trial-and-Error
Tencent's famous 'blockbuster' products often result not from top-down decisions but emerge through internal competition:
Internal Competition:Tencent encourages moderate redundancy and internal competition. The most classic case is the birth of WeChat—three teams were simultaneously working on mobile social products, and ultimately Zhang Xiaolong's Guangzhou team prevailed. While this mechanism may seem resource-intensive, it greatly reduces the risk of missing significant innovations.
Tolerance for Failure:The company culture encourages trial-and-error and does not punish employees for a single instance of poor performance. This sense of security empowers employees to take on highly challenging innovation projects.
3. Talent Management: Fluidity Mechanism and Nudges
In terms of personnel management, Tencent emphasizes 'caring for employee growth' by designing mechanisms to stimulate self-motivation:
Internal Mobility Program:Employees are free to transfer internally to find a more suitable role. This not only allows talent resources to flow but also helps employees 'get back on track' after setbacks, enhancing organizational resilience.
Nudging Rather Than Forcing:Management policies are seldom implemented in a 'one-size-fits-all' manner. Instead, they follow the 'Nudge' theory, guiding employees to make correct decisions by optimizing the environment while respecting the complexity of human nature and business operations.
Flat Management Structure:The management hierarchy is kept to within four levels to ensure efficient and transparent communication.
4. Cultural Core: User-Centric and Integrity
At the cultural level, Tencent has a distinct engineering DNA and strong value principles:
User value as the core principle:This is the heart of Tencent's business philosophy. Whether it’s the product managers’ '10/100/1000 Rule' (conducting 10 user surveys, following 100 blogs, and collecting 1,000 feedbacks each month) or senior-level decisions, all emphasize starting from user needs, even if it means sacrificing short-term commercial interests to maintain user experience.
The bottom line of integrity:In Tencent's values, 'integrity' is a high-voltage line. Zero tolerance is shown towards corruption and fraud, while high tolerance is maintained for non-principal issues (e.g., employee personalities or attire), creating an atmosphere that protects creativity despite seeming loosely structured.
It is this combination of 'front-end competition, back-end support systems, and cultural safeguards' that allows Tencent to innovate with the agility of a small company while possessing the solid technical foundation of a large enterprise.
"Lishi Business Review":What is the relationship between such management and cultural systems and the founding team of the company?
Yuanbao App:Tencent's unique management and cultural system has an inseparable, deeply imprinted relationship with its founding team. It can be said that Tencent's 'genes' are shaped by its founding team.
This is mainly reflected in the following aspects:
1. The origin of the 'Product Manager' mindset
Ma Huateng himself is an exceptionally talented 'Chief Experience Officer' and 'Super Product Manager.' His relentless pursuit of perfection in product details (such as personally replying to user emails and testing products late into the night during the early days) set a company-wide benchmark of 'placing user value above all else.' This top-down reverence for product experience directly led to specific cultural norms like the '10/100/1000 Rule.'
2. A 'Gentle yet Resolute' pragmatic approach
The founding team (especially Ma Huateng) demonstrated a pragmatic, low-key, and non-flashy working style that profoundly influenced the company's decision-making model. They preferred 'small, quick steps with trial and error iterations' over aggressive high-stakes gambles. This approach fostered the 'internal horse racing' mechanism—allowing multiple routes of exploration within controlled boundaries, ultimately letting the market choose the best solution rather than having top management impose decisions.
3. Technical background and engineering culture
Most of the founding team came from technical backgrounds, establishing a strong technological foundation for the company. Understanding the importance of a technical platform for scaling operations, they strongly supported the development of TEG (Technology Engineering Group), ensuring the accumulation and sharing of technical capabilities. Their technical expertise also made them more inclined to rely on data-driven decision-making rather than pure intuition.
4. The evolution of 'Family Culture' and talent philosophy
In its early days, the founding team at Tencent emphasized a 'family culture,' fostering a warm and supportive atmosphere. As the company grew, this evolved into a more modern approach focused on 'supporting employee growth.' The founding team's respect and tolerance for talent (such as allowing internal mobility and accepting failures) enabled the company to attract and retain a large pool of top talent, providing fertile ground for 'internal entrepreneurship.'
5. An open mindset with 'Half a Life'
Founders like Ma Huateng realized early on that Tencent could not dominate every field. This clear understanding led to the strategy of 'connecting everything' and entrusting 'half of its life' to partners through an open ecosystem. This strategic focus allowed Tencent to concentrate on its core social platforms while building an extensive external ecosystem through capital connections.
In summary:Tencent's founding team is not only the company's creator but also the 'chief architect' of its management philosophy and cultural values. They embedded their traits (pragmatism, technological sensitivity, user orientation, open mindset) into the fabric of the company and institutionalized them through system designs (such as horse racing, middle platforms, and talent mobility), allowing this system to operate independently of any individual and sustain Tencent's vast business operations.
"Lishi Business Review":How to view the future development of Tencent?
Yuanbao App:Based on Tencent's 2025 financial performance and its latest strategic moves, Tencent’s future development can be defined as a transformation from an internet giant reliant on 'traffic dividends' to a digital infrastructure engine driven by 'AI technology'.
Below is a detailed analysis based on data and strategy:
1. Core driving force: AI reshapes the entire business landscape
Tencent's future growth will no longer rely solely on user base expansion but instead depend on enhancing the 'value content' of existing businesses through AI technology.
AI Industrialization:Tencent is leveraging AI to improve game development efficiency (e.g., NPC intelligence) and ad targeting precision. The 17% year-on-year increase in advertising revenue in Q4 2025 was largely attributable to AI-driven ad targeting capabilities.
AI Ecosystem:This is the most anticipated area for Tencent's future. Tencent plans to build an 'AI-driven ecosystem' leveraging WeChat’s infrastructure, integrating intelligent capabilities into mini-programs, social interactions, and payment functionalities. In internal discussions, Ma Huateng mentioned that future AI applications will exhibit a 'human-like presence' and memory abilities, fundamentally transforming how users interact with WeChat.
Research and development investment:In 2025, Tencent’s R&D investment reached 85.75 billion yuan, with capital expenditures hitting 79.2 billion yuan, primarily allocated towards GPUs and AI infrastructure, paving the way for large-scale models like HunYuan and future applications.
2. Analysis of Growth Drivers
Tencent’s three core businesses each show clear growth trajectories:
Gaming (Stable Foundation):The domestic market remains stable thanks to long-standing titles like Honor of Kings and Peacekeeper Elite. The international market has become a new growth engine, with international gaming revenue growing by 33% year-on-year in 2025 (on a fixed exchange rate basis), driven by strong performances from PUBG MOBILE and games under Supercell.
ToB (Tencent Cloud Profit Turning Point):This marks a significant turning point. Tencent Cloud achieved full-year profitability at scale in 2025 (with adjusted operating profit reaching approximately 5 billion yuan), ending years of losses. This success is attributed to high-value-added demand for AI computing power (IaaS) and intelligent solutions (PaaS/SaaS).
Advertising (Monetization of Channels):Video accounts are accelerating monetization by optimizing recommendation algorithms and ad load rates through AI, becoming a primary driver of advertising revenue growth.
3. Financial Health and Shareholder Returns
Tencent has an exceptionally strong ability to generate cash flow, which provides a safety net for its 'aggressive' investments in the AI field.
Performance Data:Full-year revenue for 2025 reached 751.77 billion yuan (up 14% year-on-year), with net profit attributable to shareholders at 224.84 billion yuan (up 15.85% year-on-year). Gross profit margin improved to 56.21%, demonstrating strong profitability.
Shareholder returns:Tencent plans to distribute a dividend of HKD 5.30 per share, totaling approximately HKD 48 billion, while continuing stock repurchases, reflecting management's confidence in future cash flows.
4. Potential Risks and Challenges
Short-term Profit Margin Pressure:To secure a leading position in AI, Tencent's AI-related investments for 2026 are expected to double (exceeding 36 billion yuan), which could pressure Non-IFRS operating profit margins in the short term.
Macroeconomic Environment:The fintech business (payments, credit) is strongly correlated with macroeconomic activity. If consumer recovery falls short of expectations, growth in this segment may be constrained.
5. Institutional Perspectives
The market is optimistic about Tencent's future. As of April 2026, approximately 95% of the covering institutions have assigned a 'Buy' or 'Outperform' rating, with an average target price for the Hong Kong shares at around HKD 733.25, considering Tencent as 'the most certain core asset in China’s digital economy'.
Summary:Tencent's future lies in‘AI in All’It is no longer just a social networking or gaming company but is attempting to transform its vast social, content, and transaction data into intelligent services through AI technology, building an ecosystem at the level of an 'intelligent operating system'.
Lishi Business Review:How to evaluate the differences between the latest large model released after Yao Shunyu joined Tencent and its previous versions?
Yuanbao App:After Yao Shunyu joined Tencent, the HunYuan large model team conducted a 'complete overhaul' of the underlying logic and R&D system, launching the brand-new HunYuan Hy3 preview. Compared to previous versions, this update represents not only a technical parameter upgrade but also a complete shift in R&D philosophy.
1. R&D Philosophy: From 'Parameter Competition' to 'Practical Focus'
Under the leadership of Yao Shunyu, the team established three core principles that directly altered the development direction of the model:
Systematization of capabilities:Rejecting 'one-sidedness', no longer blindly pursuing leaderboard scores, but emphasizing the deep integration of various capabilities such as reasoning, coding, and tool utilization to meet the complex demands of real-world scenarios.
Authenticity of evaluation:Proactively moving away from publicly available datasets that are easy to 'game', focusing on self-created questions and manual evaluations to examine the model's 'combat effectiveness' in the real world.
Cost-performance pursuit:Through the deep integration of model architecture and inference frameworks, significantly reducing task costs and making the model more advantageous for commercial implementation.
2. Technical Architecture: Fusion of Fast and Slow Thinking
This is the most prominent technical feature of Hy3 preview. The previous generation HunYuan 2.0 used two separate versions, Think (deep thinking) and Instruct (quick response), requiring users to switch manually.
Single-model fusion:Hy3 preview achieves the integration of fast and slow thinking. The model can automatically adjust its depth of thought based on the complexity of the task, allowing users to receive responses that balance efficiency and quality without any intervention.
Parameter optimization:The total number of parameters has been streamlined from 406 billion to 295 billion, with active parameters reduced to 21 billion. This 'subtractive' strategy effectively avoids the quagmire of simply stacking parameters and improves reasoning efficiency.
Core capabilities: Focus on Agent and code
The new model takes Agent capability as a core breakthrough point, significantly enhancing performance in complex task decomposition, tool invocation, and long-context understanding.
Performance improvement:In benchmark tests such as SWE-Bench Verified (code generation), Hy3 preview's performance is approaching GLM-5 and Kimi-K2.5, achieving leading domestic results in hardcore reasoning tasks like the mathematics doctoral qualifying exam.
Scenario implementation:The model has been integrated into dozens of products such as Tencent Y元宝, Tencent Docs, and QQ, showcasing more natural and immersive interaction experiences in scenarios like AI NPCs in 'Peacekeeper Elite.'
Organization and infrastructure: Integrated research and operations
Yao Shunyu broke the previous separation between infrastructure and algorithm research by merging AI Infra and the large language model department into one unit.
This 'research-engineering integration' approach in the OpenAI style has enabled the team to efficiently roll out this flagship model just two months after rebuilding the pre-training and reinforcement learning infrastructure.
The Lishi Business Review:Compared to the latest large models from Alibaba's Qwen and ByteDance's DouBao, is Tencent's HunYuan large model leading or lagging behind?
Yuanbao App:Tencent's HunYuan large model is currently in the phase of catching up and gradually approaching the top tier. Compared to Alibaba's Qwen and ByteDance's DouBao, it does not lead comprehensively across all benchmarks but demonstrates a 'practicality-oriented, scenario differentiation' characteristic.
Core Capability Comparison:
Mathematics and Reasoning:HunYuan Hy3 preview performed impressively, scoring the highest domestic mark (88.4 points) in the Tsinghua University Qiuzhen Academy Mathematics Qualifying Exam and achieving an 84.3% score on the IMO-AnswerBench, second only to DeepSeek and slightly higher than Qwen.
Code and Agents:Compared to DouBao Seed 2.0 Pro and Qwen 3.6 Max, HunYuan still lags behind in code generation and complex agent tasks, but it has surpassed Kimi K2.5 in real-world agent implementation capabilities.
Multimodal and Ecosystem:ByteDance's DouBao excels in multimodal (image/video) and content creation ecosystems; Alibaba's Tongyi Qianwen is more specialized in long-text processing and office scenarios; Tencent HunYuan is deeply integrated into the WeChat/QQ ecosystem, making it most convenient for social and office use cases.
Overall evaluation:The HunYuan large model does not aim to be the best at everything but focuses on 'comprehensive practicality' and cost-effectiveness, with a key emphasis on enhancing synergy with Tencent products such as Tencent Docs and Yuanbao. Although its peak performance is slightly behind DouBao and the top versions of Qianwen, it has firmly secured its place in the first tier domestically and is fully capable of supporting Tencent’s core business needs.
"Lishi Business Review":Do you think Yao Shunyu has proven his capabilities through HunYuan Hy3 preview?
Yuanbao App:Undoubtedly, Yao Shunyu has effectively demonstrated his abilities using HunYuan Hy3 preview. This is not just a version update of the model but also the first hardcore result delivered under his leadership in Tencent's 'HunYuan Reconstruction Plan.'
In terms of outcomes, his performance has primarily manifested in breakthroughs across several key dimensions:
1. Addressing weaknesses and returning to the game
The most visible achievement of Hy3 preview is significantly narrowing the gap between Tencent HunYuan and leading domestic large models such as Kimi and GLM. In critical metrics like complex reasoning (e.g., achieving the highest score domestically in Tsinghua University's mathematics doctoral qualification exam), code generation (SWE-Bench Verified reaching 74.4%), and intelligent agent (Agent) capabilities, the new model has shown remarkable improvements, proving his ability to lead the team in resolving previous foundational issues where HunYuan fell short.
2. Correcting the R&D Path: From 'Benchmark Chasing' to 'Practical Use'
Yao Shunyu demonstrated his strategic vision as the chief scientist. He keenly pointed out the issues caused by HunYuan's previous excessive focus on benchmark rankings, which led to data contamination and weak real-world capabilities. He also spearheaded a fundamental shift in the R&D logic:
Rejecting specialization bias:No longer blindly pursuing parameter scale and public leaderboard rankings; instead, emphasizing systematic collaboration across various capabilities such as reasoning, coding, and long-text processing.
Real-world evaluation:Actively stepping away from datasets prone to 'benchmark gaming,' and establishing over 50 proprietary evaluation systems to focus on assessing the model’s 'combat effectiveness' in real-world scenarios.
3. Implementation Results and Efficiency Optimization
Another measure of capability is whether technology can be transformed into productivity. Hy3 preview has not only been rapidly integrated into more than ten core products within Tencent, such as Yuanbao, Tencent Docs, and CodeBuddy, but also achieved a 40% improvement in reasoning efficiency while significantly reducing API invocation costs (minimum input cost of 1.2 yuan per million tokens). This pursuit of cost-effectiveness reflects his control over the rationality of commercial implementation.
4. Restructuring Organizational Architecture
Beyond the technology itself, he has also proven his leadership in organizational management. After taking office, he swiftly restructured the AI Infra (Artificial Intelligence Infrastructure) team, streamlining the entire chain from pre-training, fine-tuning, to reasoning, breaking down past departmental barriers. This 'research-engineering integration' approach forms the foundation for HunYuan’s rapid 'reconstruction' within a short period (less than three months from training to launch).
In summaryAlthough Hy3 preview still lags behind the world's top closed-source models (such as Claude or GPT) in some extreme tasks, Yao Shunyu has successfully demonstrated his comprehensive ability to identify pain points, correct directions, execute efficiently, and deliver high-value results.
Lishi Business Review:How to predict Tencent's next evolution in the large model field?
Yuanbao App:Predicting Tencent’s next evolution in the large model field can be summarized as 'technology returning to practical use, ecosystem becoming fully intelligent.' Based on the latest developments in April 2026 (such as the release of Hy3 preview) and financial report data, Tencent's evolutionary path is becoming very clear.
1. Model layer: Transition from 'benchmark chasing' to 'real-world application'
Under Yao Shunyu’s leadership, Tencent HunYuan's large model will completely move away from a pure parameter competition towards 'fast-slow thinking integration' and 'high cost-performance.'
Systematic capabilities:Rather than pursuing extremes in single metrics, there will be an emphasis on deep synergy across reasoning, coding, long text, and other abilities. Hy3 preview has already shown significant improvements in complex logical reasoning and agent (Agent) tasks, making future models more like versatile 'digital employees' rather than simple conversational bots.
Reject vanity metrics:Tencent will focus more on the performance of models in real business scenarios (such as Tencent Docs, Yuanbao) rather than public Benchmark rankings. This 'pragmatism' will help HunYuan build a differentiated advantage in specific productivity scenarios (such as code generation, data analysis).
2. Application layer: The 'Agentization' of super entry points
Tencent's biggest moat - WeChat and QQ - will complete the evolution from 'connector' to 'intelligent hub.'
Decentralized agent ecosystem:In the future, when you open WeChat, it won't just be for chatting. Instead, through 'Yuanbao' or direct dialogue, you'll invoke various Mini-program Agents to accomplish complex tasks like ordering food, booking tickets, and managing schedules. Tencent is working to make every mini-program capable of AI-driven autonomous execution.
Full-scenario penetration:Beyond social networking, Tencent Cloud will launch an Agent development platform covering the entire chain (such as WorkBuddy, CodeBuddy), encapsulating large model capabilities into 'building blocks' that enterprises can access with low barriers, accelerating implementation in B-end scenarios such as finance, healthcare, and mobility (e.g., smart cockpits).
3. Strategic layer: Dual drivers of infrastructure and open source
Increased investment in computing power infrastructure:To support model iteration, Tencent's capital expenditure in 2025 has reached 79.2 billion yuan, with R&D investment at 85.75 billion yuan. In the future, Tencent will continue to heavily invest in AI Infra (infrastructure) to ensure sufficient 'fuel' for rapid iteration of the HunYuan large model.
Embrace the open-source community:The open-sourcing of Hy3 preview is just the beginning. To make up for past delays in the developer ecosystem, Tencent will be more proactive in building influence through open source in the future, leveraging community power to enhance model optimization and forming a closed loop of 'application-feedback-iteration'.
In summary:Tencent’s next phase of evolution will no longer focus on aggressively competing in 'parameter scale' but instead utilize a combination of 'Hybrid Model + Super Entry Point + Engineering Implementation Capability' to transform AI into an infrastructure as omnipresent as 'tap water.'
Risk Disclaimer: The above content only represents the author's view. It does not represent any position or investment advice of Futu. Futu makes no representation or warranty.Read more
Comments
to post a comment