Chinese technology major Tencent has entered another phase of the rapidly developing artificial intelligence competition with the release of its new Hy4 preview model.
Tencent announced on August 28, 2026, that it had released and open sourced Hy4 preview, describing it as a next generation large language model focused on real world productivity. The model has been designed for a broad range of applications, including software engineering, office work, financial analysis, game development and scientific research.
The release comes as competition among Chinese artificial intelligence companies continues to intensify. Companies such as Z.ai and Moonshot AI have also been developing increasingly capable open and open weight models, creating a highly competitive domestic AI market.
Tencent said Hy4 preview has 770 billion total parameters, with 49 billion parameters activated for each token. It also supports a context window of more than one million tokens. The architecture is based on a mixture of experts approach, which allows the model to activate a selected portion of its parameters for individual tasks rather than using the entire model for every operation.
One of the key highlights of the launch is Tencent’s internal evaluation of Hy4.
The company conducted a blind side by side assessment involving 163 internal experts who evaluated model responses across 203 engineering tasks. Hy4 preview achieved an average score of 2.99 out of 4.00.
In the same evaluation, GLM 5.3 from Z.ai received an average score of 2.92, while Kimi K3 from Moonshot AI scored 2.94.
These results placed Hy4 slightly ahead of the two competing models in Tencent’s internal assessment. However, the results should not be interpreted as an independent industry benchmark because the evaluation was conducted internally by Tencent using its own experts and testing framework.
The distinction is important because artificial intelligence models can perform differently depending on the benchmark, tasks, scoring methodology and evaluation group.
Tencent has positioned Hy4 primarily as a productivity focused model rather than a system built only for general conversational use.
In software engineering, Tencent said Hy4 has improved its ability to understand long term development tasks, create plans, identify and fix coding problems and validate results. The model has also been designed to improve front end development, including the visual quality and interaction of applications.
For office and analytical work, Hy4 is designed to process information spread across multiple files and turn that information into useful outputs. Tencent said the model can assist with data analysis, financial models, documents, spreadsheets and presentations.
Financial analysis is another major focus area.
The model has been trained using data and expertise from Tencent professionals working in areas such as finance, software engineering, gaming and security. The company said this approach was intended to improve performance on practical workplace tasks rather than focusing solely on conventional academic benchmarks.
Hy4 also has applications in game development.
According to Tencent, the model can generate a playable game prototype from a natural language request and work with game engines. Developers can then continue modifying and expanding the project through multiple rounds of interaction.
This could make AI increasingly useful as a development assistant for smaller teams and individual developers, although the practical quality of generated software and game projects will depend on the complexity of the task and the level of human supervision.
Scientific research is another area where Tencent says Hy4 has made progress.
The company highlighted improvements in understanding, reasoning and problem solving for complex research questions. Areas mentioned include artificial intelligence research, molecular dynamics, condensed matter physics and fundamental mathematics.
Tencent has also integrated Hy4 with several of its existing products.
The model can be accessed through Tencent platforms including WorkBuddy, CodeBuddy, Yuanbao and ima. Tencent has also made the model available through APIs, allowing developers and businesses to integrate it into their own applications.
The company said Hy4 would be available for a limited free trial on WorkBuddy and CodeBuddy for two weeks following its launch.
The open source release is particularly significant because developers and researchers can examine and deploy the model under its published licensing terms rather than relying exclusively on a closed commercial service.
Tencent has also released the model through platforms including Hugging Face and ModelScope. The official model repository provides deployment information and technical documentation for developers.
The company says Hy4 is designed to improve efficiency as well as intelligence.
Tencent reported that the model participated in parts of its own development and optimisation process. According to the company, Hy4 helped explore training methods, data strategies, evaluation approaches and low level optimisation. Tencent also said the model identified bottlenecks in its inference system and contributed to optimisation work.
The company reported a 31.8 percent improvement in end to end inference throughput compared with its baseline in the optimisation process. These are Tencent's own reported results and have not been independently verified.
Despite the strong claims, Tencent acknowledges that Hy4 preview is an early version.
The company's technical documentation states that the model still has room for improvement in both pre training and post training. It also identifies some limitations, including spending too much time reasoning through certain complex tasks and a tendency to over verify its own answers.
This means users should not assume that Hy4 will outperform every competing AI model across every type of task.
The broader significance of Hy4 lies in the accelerating competition among Chinese AI developers.
China's technology companies have been releasing advanced models at a rapid pace, competing in areas such as reasoning, coding, multimodal capabilities, enterprise applications and open source development.
Tencent's latest release adds another major technology company to that competition.
The model also demonstrates how AI development is increasingly moving beyond general chatbots. Companies are now focusing heavily on specific workplace applications where AI can directly assist with coding, data analysis, document creation, research and business operations.
For businesses, these capabilities could reduce the time required for repetitive analytical and technical work. However, human review remains important, particularly for financial analysis, scientific research, software development and other areas where incorrect AI outputs can create significant consequences.
The competition between Hy4, GLM models from Z.ai and Kimi models from Moonshot AI is likely to continue as each company releases new versions and improves its systems.
Tencent's internal testing provides an early indication that Hy4 is competitive with some recent Chinese models, but broader independent evaluations will be needed to determine how it compares across different tasks and benchmarks.
For now, Tencent's release marks another significant development in China's artificial intelligence sector.
With 770 billion total parameters, 49 billion active parameters and a context window exceeding one million tokens, Hy4 represents a substantial investment in large scale AI infrastructure and model development.
Its open source availability also gives developers around the world an opportunity to experiment with the model and assess its capabilities independently.
The next stage of the competition will depend not only on benchmark scores but also on how effectively these models perform in real world environments.
Tencent has indicated that additional models in the Hy4 series are expected to follow. As development continues, the company is likely to use feedback from users and its own products to improve the model further.
For the global AI industry, the release reinforces a broader trend in which open models are becoming increasingly capable and competition is expanding beyond the traditional leading technology companies.
Tencent's Hy4 preview is therefore an important new entry in China's rapidly evolving AI landscape, with its strongest focus currently placed on practical productivity, coding, business analysis and research applications.