Today, we are launching the Hy4 Preview. It features a total of 770 billion parameters, with 49 billion active parameters and a context length of 1 million tokens. The model demonstrates exceptional capabilities in real-world productivity tasks such as coding, office work, and scientific research.
Next-Generation Flagship Model
The Hy4 Preview has seen significant expansions in model size, context length, and data scale. Joint advancements in pre-training and post-training have driven another major leap in intelligence levels, firmly establishing it among the top tier of open-source models.
Built for Productivity
Through the co-creation of high-quality data with top experts from Tencent's internal software engineering, gaming, finance, and security divisions, the Hy4 Preview has achieved significant progress in various real-world productivity tasks:
Software Engineering:Enhanced capabilities in understanding, planning, debugging, and verification for long-horizon development tasks, further improving visual aesthetics and interaction quality in front-end development;
Office Analytics:Significantly enhance understanding of complex office environments and financial analysis capabilities, with a focus on optimizing data analysis and cross-file collaboration to complete the entire workflow from information processing to the delivery of documents, spreadsheets, and presentations;
Game Development:Enhance the ability to directly generate playable prototypes from single-sentence requirements, and enable proficient use of game engines, allowing developers to continuously refine complex game projects through multi-round interactions;
Scientific Research:Significantly improve the understanding, reasoning, and problem-solving capabilities for complex scientific research issues. The model has made substantial progress in various scenarios such as AI R&D, molecular dynamics simulations, condensed matter physics, and fundamental mathematics.
Meanwhile, H...

On August 12, Tencent (00700.HK) released its second-quarter financial report, with AI serving as the core theme throughout the quarter. From completing reconstruction and launching the Hy3 preview in under three months, to rapidly iterating to the official Hy3 release, and with the larger-scale Hy4 model scheduled for imminent launch, Hunyuan is developing stable and systematic model R&D capabilities. Tencent has established an initial advantage in the office AI agent sector, with multiple products gaining momentum simultaneously. The WeChat AI assistant "Xiao Wei" launched in June and continues to expand its limited beta testing scope.
Tencent's fundamentals remained robust this quarter: Revenue reached RMB 204.79 billion, up 11% year-on-year; Non-IFRS operating profit was RMB 75.64 billion, up 9% year-on-year. Excluding the impact of revenue, costs, and expenses from new AI products (primarily Hunyuan, Yuanbao, CodeBuddy, WorkBuddy, and Xiao Wei), Non-IFRS operating profit increased 19% year-on-year to RMB 86.1 billion.
A sound business model and strong fundamentals support more substantial AI investment: R&D expenditure for the quarter was RMB 27.28 billion, up 35% year-on-year; capital expenditure was RMB 52.78 billion, up 176% year-on-year and 65% quarter-on-quarter. Excluding prepayments for computing power procurement, free cash flow stood at RMB 37.6 billion. As AI contributes more incremental growth to core businesses, the positive cycle between investment, implementation, and business returns is gradually strengthening.
Ma Huateng, Chairman and CEO of Tencent, stated: ...

On July 6, Tencent officially launched HunYuan Hy3. Compared to the preview version, Hy3 demonstrates significantly stronger performance than models of similar size and matches the intelligence level of flagship models with 2–5 times more parameters, while further reducing pricing and substantially improving overall stability and cost-effectiveness. Hy3 has already been integrated into multiple business applications including WorkBuddy/CodeBuddy, Yuanbao, Marvis, and IMA. Its API is now available on Tencent Cloud TokenHub, with several international API platforms set to onboard it soon.
Model Intelligence Leaps Forward Across the Board; Agent Practicality Achieves a Qualitative Breakthrough
Hy3 is a model that integrates fast and slow thinking, built on a Mixture-of-Experts (MoE) architecture with 295 billion total parameters and 21 billion activated parameters, supporting a context length of up to 256K tokens. The Hy3 preview released on April 23 marked the first version after HunYuan’s reconstruction, achieving qualitative improvements over Hy2 in complex reasoning, instruction following, in-context learning, code generation, and agent capabilities. Hy3 continues along a steep and clear trajectory of capability growth, further enhanced by scaling up post-training compute resources and improving data quality and diversity. As a result, Hy3 now outperforms its preview version across various tasks and, for the first time with a relatively compact size, matches the performance of large-scale flagship models both domestically and internationally.
Hy3 shows particularly notable improvements in productivity-oriented tasks such as software development, office productivity, financial modeling, front-end design, and game production, making it a highly cost-effective and reliable choice...

Just now, Tang Daosheng, Senior Executive Vice President of Tencent and CEO of the Cloud and Smart Industries Group, held a dialogue with Yao Shunyu, Tencent’s Chief AI Scientist, at the China National Convention Center in Beijing, discussing Tencent’s thinking and progress on large models and AI products.
Below is the full transcript of their conversation:
What is the 'first principles' approach to building models versus building products?
Tang Daosheng: We’re very glad to have you here, Shunyu.
Yao Shunyu: Hello everyone. I usually stay in Haidian District and rarely come to Chaoyang District—glad to be here.
Tang Daosheng: Our conversation today may take a rather novel format, and if anything unexpected happens, I hope it’ll be a pleasant surprise for everyone.
Shunyu, before you joined Tencent, I remember asking you some questions: Why did you choose to join Tencent for the 'second half'? And what do you think matters most in the second half of AI?
Yao Shunyu: First, let me clarify what 'second half' means. Lately, I feel this term has been somewhat overused—it actually originated from a blog post I wrote last year. What does it mean? In my view, AI had already been developing for decades prior to last year, but the greater focus was on solving problems and finding good methods. Recently, however, it’s become clear that methodologies have matured significantly, while identifying the right problems has grown more challenging.
For example, in the past we developed methods like AlphaGo to play Go, but such methods were only suitable for Go or other board games. You might build a specialized model for translation, but it could only perform translation and nothing else.
...

In the next phase of AI, intelligence is shifting from 'product capability' to 'productivity.' This conference, themed 'Agents Entering the Field, Driving Efficiency and Growth,' will unveil an upgraded Tencent Cloud Agent strategy and product portfolio, and share methodologies and latest practices for enterprise-grade Agent implementation.

At noon today, a self-media outlet falsely reported that 'the head of AI is about to resign,' insinuating Tencent. This rumor is completely baseless.
Tencent hereby issues a serious clarification and reserves the right to pursue legal responsibility against those who maliciously spread this rumor.

On May 13, Tencent (00700.HK) released its Q1 financial report, showing a comprehensive acceleration in AI development: model reconstruction and release, new cloud launches, the introduction of an Agent product matrix, and AI integration into existing products have all advanced simultaneously. In just a few months, Tencent has achieved dozens of significant updates and releases in AI, primarily focused on the Agent ecosystem and cloud-based infrastructure.
Ma Huateng, Chairman of the Board and Chief Executive Officer of Tencent, stated: 'We have made remarkable breakthroughs in new AI products and continue to empower core business growth with AI. The reorganized AI R&D team has rebuilt our AI infrastructure and developed the Hy3 preview model, which leads in performance among models of similar parameter scale while offering both practicality and cost-effectiveness. Since April 28, it has consistently ranked at the top of OpenRouter’s token consumption list. Our AI efficiency agent solutions are already showing results, with WorkBuddy currently being the most widely used AI efficiency agent service in China. Meanwhile, our core businesses continue to grow in user stickiness, revenue, and profitability, providing ample cash flow support for AI investment and laying a rich foundation for AI application scenarios.'
Benefiting from the accelerated AI strategy, Tencent's Q1 revenue reached 196.46 billion yuan, a year-on-year increase of 9%; Non-IFRS operating profit was 75.63 billion yuan...

On April 23, the Tencent HunYuan Hy3 preview language model was released and open-sourced. This is a hybrid expert model that integrates fast and slow thinking, with a total of 295 billion parameters, 21 billion activated parameters, and supports a maximum context length of 256K tokens. It is the first model trained after HunYuan’s reconstruction and also the smartest model developed by HunYuan to date. It has achieved significant improvements in complex reasoning, instruction following, contextual learning, coding, agent capabilities, and inference performance.
In February 2026, Tencent HunYuan rebuilt its pre-training and reinforcement learning infrastructure, along with three principles for pursuing model practicality:
1. Systematized capabilities: No preference for specialization, as even a single application such as a code agent involves deep collaboration across various capabilities like reasoning, long-form text, instructions, dialogue, coding, tools, and more.
2. Authentic evaluation: Actively stepping away from public leaderboards that are easily 'gamed,' assessing and improving the model’s real-world effectiveness through self-created questions, the latest exams, human evaluations, product beta tests, and other methods.
3. Cost-performance balance: Practicality cannot be separated from commercial viability. By deeply integrating model architecture and inference framework design, the cost of tasks has been drastically reduced, making intelligence both affordable and effective.
Hy3 preview can be seen as the starting point for HunYuan’s rapid exploration of large-scale practical models aimed at solving real-world problems.
Tencent's Chief AI Scientist, Shunyu Yao, stated that Hy...