Tencent's new generation of large model Hy4 will be released and will become the core engine of WorkBuddy. Following the public notice at the previous financial report conference, the internal test interface of Tencent's new generation of large model Hy4 has been officially exposed recently, which means that its official release is not far away.

Tencent's new generation of large model Hy4 will be released and will become the core engine of WorkBuddy. Following the public notice at the previous financial report conference, the internal test interface of Tencent's new generation of large model Hy4 has been officially exposed recently, which means that its official release is not far away. Judging from the leaked model selection interface, Tencent has continued its established strategy of "paralleling self-developed models and integrating excellent open source solutions" and provides a total of three core options: Hy4, which is positioned as an expert-level large model, Hy3, which is marked as a new upgrade and suitable for most scenarios, and DeepSeek, which focuses on reasoning and understanding. As the focus of this upgrade, the overall positioning of Hy4 is significantly higher than that of the previous generation Hy3, and it is endowed with expert-level capabilities to solve more complex business and technical problems. On the path of technology evolution, the industry generally speculates that the substantial improvement in performance may come from two directions: one is to directly move the parameter scale to the 1 trillion level to tie the industry's flagship level; the other is to further strengthen RL (reinforcement learning) post-training based on the existing Hy3 architecture, thereby significantly enhancing the core intelligence limit without blindly expanding parameters. Looking back at the Hy3 large model launched this year, its total parameter volume is 295B (activation parameters 21B, MTP layer parameters 3.8B), and it performs well in high-demand tasks such as software development, office production, financial modeling, front-end design, and game production. In a large-scale expert real-work blind test organized in-house, the overall performance of Hy3 was even better than that of similar models. As the Hy4 with higher positioning and stronger performance is about to debut, it is expected that it will fully take over and become the core underlying support for the intelligent agent application WorkBuddy, further unleashing Tencent's application potential in AI intelligent agent implementation and productivity scenarios. via AI News (author: AI Base)