China's AI race shows little sign of slowing down, with Tencent becoming the latest major tech company to introduce a new model. The Chinese technology giant this week released a preview of Hy4, an open-source AI model aimed at practical applications including coding, financial analysis, office tasks and scientific research.
Tencent claims Hy4 narrowly outperformed recent models from Chinese rivals Z.ai and Moonshot AI in an internal evaluation. In a blind test involving 163 experts and 203 engineering tasks, Hy4 achieved an average score of 2.99 out of 4.00. Z.ai's GLM-5.3 scored 2.92, while Moonshot AI's Kimi K3 recorded 2.94. However, these results come from Tencent's own testing and have not been independently verified.
The launch comes as Chinese AI companies continue to release increasingly capable open models at a rapid pace. Z.ai and Moonshot AI, in particular, have been challenging the dominance of US-based AI companies in areas such as reasoning and coding.
Moonshot recently introduced Kimi K3, which it described as a 2.8-trillion-parameter open-weight model built for advanced reasoning, programming and other complex tasks. Z.ai has also attracted attention with its latest GLM models, including GLM-5.3-Flash, which initially appeared under the name Ox Alpha before the company confirmed its involvement.
WHAT CAN TENCENT'S HY4 DO?
On paper, Hy4 is significantly smaller than Kimi K3, but Tencent says it represents a substantial improvement over its previous generation. The model has 770 billion total parameters, with around 49 billion activated for an individual request. It also supports a context window of more than 1 million tokens, allowing it to process very large amounts of information in a single interaction.
Tencent says Hy4 has been developed with practical, real-world applications in mind rather than being optimised solely for benchmark performance.
For software development, the model can assist with understanding projects, planning, debugging and validating code. It can also be used for financial and data analysis, generating documents and presentations, and creating playable game prototypes based on natural-language instructions.
Tencent claims Hy4 has also improved its performance in scientific applications, including mathematics, AI research and molecular dynamics.
HY4 ALSO HELPED DEVELOP ITSELF
One of the more notable aspects of Hy4 is Tencent's claim that the model played a role in its own development.
According to the company, Hy4 was used in experiments involving training techniques, data strategies and evaluation systems. The results from those experiments were then used to guide additional rounds of optimisation.
Tencent refers to this as an early-stage “recursive self-improvement loop”. In simple terms, the model was not merely trained by researchers; it was also used to test different ways of training and improving AI systems, with those findings feeding back into the development process.
WHERE CAN YOU USE TENCENT HY4?
Tencent has released Hy4 as an open-source model, with access available through several of its products, including CodeBuddy, WorkBuddy, Yuanbao and ima. Developers can also access the model through APIs.
For a limited period, Tencent is offering free access to Hy4 through WorkBuddy and CodeBuddy for two weeks.
