WorkBuddy's perfect partner: Tencent's next-generation large model Hy4 is about to be released, featuring expert-level capabilities.

At a recent earnings call, Tencent teased the upcoming release of its next-generation large model, Hy4. Now, a suspected internal interface for the model has leaked, suggesting the launch is imminent.

In the model selection interface, Tencent offers three options — Hy4, labeled as an expert-level large model, followed by Hy3, which carries a "new upgrade" badge and is described as a brand-new model suited for most scenarios.

There is also a DeepSeek option, described as focused on reasoning and understanding.

This aligns with Tencent's previous strategy: in addition to deploying its own large models, it also integrates excellent open-source models like DeepSeek into its services.

Based on the descriptions, the Hy4 large model is positioned higher than Hy3, targeting expert-level tasks and more complex problems. This implies a significant performance boost for Hy4, though no specific data on the improvements has been released.

There are two likely directions for Hy4's enhancements. One is a larger parameter count — other companies' large models are moving toward 1 trillion parameters, and Tencent may push Hy4 beyond that threshold to improve performance.

Alternatively, Tencent could further apply RL post-training on top of Hy3, similar to how Zhipu evolved from GLM-5.2 to GLM-5.3, achieving substantial performance gains without increasing parameter size.

Tencent has developed WorkBuddy, a leading domestic agent application, where the Hy3 model performs exceptionally well. Hy4 is expected to further solidify its role as the backbone of WorkBuddy, unlocking greater potential for the application.

At the end of April this year, Tencent released the Hy3 Preview version, followed by the official Hy3 release on July 6, with a total of 295B parameters, 21B activated parameters, and 3.8B MTP layer parameters. It demonstrated significantly stronger intelligence than models of similar size, rivaling larger flagship models, and greatly enhanced practical value across various products and productivity tasks.

Hy3 showed particularly notable improvements in productivity tasks such as software development, office productivity, financial modeling, front-end design, and game development, making it a cost-effective and reliable choice.

In an internal blind test involving 270 experts evaluating models on real work tasks, Hy3 (average score 2.67/4) outperformed GLM5.1 (average score 2.51/4), with especially strong advantages in categories like front-end, data and storage, and CI/CD.