Tencent opens Hy4 preview in an open-source AI push

Tencent has released and open-sourced Hy4 preview, the latest addition to its Hunyuan AI family.

The launch places Tencent in the latest round of competition over open-source large language models, as companies seek to improve reasoning performance while limiting inference costs. The issue has become more prominent as businesses weigh whether advanced AI systems can be deployed at scale without sharply higher computing bills.

Hy4 preview uses a Mixture-of-Experts design with 770 billion total parameters and 49 billion active parameters. It supports a context window of more than 1 million tokens, allowing it to process long documents, software codebases, and large datasets in a single prompt.

Tencent is making the model available through several of its products, including WorkBuddy, CodeBuddy, Yuanbao, and ima. It can also be accessed through an application programming interface on Tencent Cloud TokenHub and OpenRouter.

Hy4 preview was developed for software engineering, office tasks, financial analysis, scientific research, and game development. For software development, the model was trained to improve long-context understanding, planning, debugging, and validation, while also supporting front-end development.

For office and analytical work, Tencent said the system performs better in complex working environments and financial analysis. It has also been tuned for data analysis and cross-document collaboration, including the production of documents, spreadsheets, and presentations.

In game development, Hy4 preview can generate a playable prototype from a single natural-language request and work with game engines through multi-turn interactions. In scientific research, Tencent said it has improved the model's performance in AI research and development, molecular dynamics simulation, condensed-matter physics, and fundamental mathematics.

The model also played a role in its own development. According to Tencent, Hy4 preview took part in automated optimisation of training methods, data strategies, evaluation frameworks, and low-level operators by proposing approaches, running experiments, and iterating on the results.

Tencent described that process as an early recursive self-improvement loop. It added that the model analysed bottlenecks in its inference system and went through several rounds of optimisation, including operator fusion and communication, raising end-to-end throughput by 31.8% from a baseline measure.

Benchmark focus

Tencent framed the release around practical use rather than benchmark scores alone, but it also published internal comparative data. In a blind evaluation involving 163 experts and 203 engineering tasks, Hy4 preview achieved an average score of 2.99 out of 4.00, ahead of GLM-5.3 at 2.92 and Kimi K3 at 2.94, according to the company.

That emphasis reflects a broader shift in the AI market. As model sizes continue to rise, developers and enterprise buyers are paying closer attention to the trade-off between reasoning quality and the cost of running models repeatedly in production systems.

Tencent said Hy4 preview was trained on high-quality data co-created with specialists in software engineering, gaming, finance, and security, as well as with the product teams behind WorkBuddy and CodeBuddy. The approach was intended to align the model more closely with real-world workflows rather than general-purpose demonstrations.

Cost pressure

Pricing is another focus. Tencent said application programming interface pricing for Hy4 preview is set at USD $0.834 per million input tokens, USD $2.501 per million output tokens, and USD $0.042 per million tokens for cache hits.

Those figures give developers an early signal on the economics Tencent expects to offer as competition intensifies among Chinese and US AI groups. Lower active parameters in a Mixture-of-Experts system can reduce the computation required for each task, even when the overall model remains very large.

Tencent also said free access to Hy4 preview will be offered on WorkBuddy and CodeBuddy for two weeks. It added that free access to Hy3 on both platforms has been extended.

The preview release is intended to gather feedback from developers and enterprise users before broader availability. Tencent added that Hunyuan follows a preview-first approach so models can be refined through real-world usage.

Hy4 preview represents a significant expansion in model size, context length, and data volume, and advances in both pre-training and post-training have led to a major leap in overall intelligence, placing it among the top tier of open-source models.