Tencent’s Next-Generation AI Model Hy4 to Launch, Set to Become WorkBuddy’s Core Engine.

After Tencent previously signaled its plans during an earnings call, a newly surfaced internal testing interface suggests that the company’s next-generation large model, Hy4, is moving closer to an official debut.

The leaked interface offers an early look at Tencent’s current model lineup and reinforces a strategy the company has already been following: combining in-house foundation models with high-performing open-source integrations, rather than relying on a single technical path.

Tencent appears to be preparing a three-model lineup

Based on the testing screen that has circulated, users are currently shown three core model options:

• Hy4 — positioned as an expert-level large model

• Hy3 — an upgraded version designed for most general scenarios

• DeepSeek — focused on reasoning and comprehension

This setup points to a layered product strategy. Instead of pushing one model to handle every use case, Tencent seems to be aligning different systems with different task profiles, balancing breadth, specialization, and deployment flexibility.

Hy4 is being positioned above Hy3

Among the three options, Hy4 clearly stands out as the centerpiece of the next phase. Its positioning is noticeably higher than that of Hy3, with Tencent framing it as an expert-grade model aimed at tackling more demanding technical and business problems.

If that direction holds at launch, Hy4 is likely intended for more complex productivity workloads and advanced enterprise scenarios, making it a major step in Tencent’s broader effort to upgrade its large-model capabilities.

Industry expectations are rising around its technical path

Although Tencent has not publicly detailed Hy4’s underlying architecture, discussion in the industry has largely centered on two possible directions for its performance gains.

2. Scaling up the parameter count, potentially pushing the model toward the trillion-parameter range in order to compete more directly with top-tier flagship models.

4. Strengthening post-training through reinforcement learning, building on the existing Hy3 foundation to raise the model’s effective intelligence ceiling without depending too heavily on brute-force parameter expansion.

At this stage, these remain outside observations rather than confirmed product details. Even so, they reflect how strongly the market expects Hy4 to deliver a meaningful leap over the previous generation.

Hy3 has already been tested in demanding scenarios

Part of that anticipation comes from the performance Tencent has already demonstrated with Hy3, which was introduced earlier this year. According to available information, Hy3 has a total parameter count of 295B, including 21B activated parameters and 3.8B MTP layer parameters.

Tencent has indicated that Hy3 has already proven capable in a range of high-requirement tasks, including:

• software development

• office productivity

• financial modeling

• front-end design

• game production

In large-scale internal blind evaluations involving expert users and real work tasks, Hy3 reportedly delivered stronger overall performance than comparable models in its category.

Hy4 is expected to become the engine behind WorkBuddy

With Hy4 now nearing release and positioned as a more capable successor, outside observers increasingly expect it to become the core foundational model behind Tencent’s AI agent application, WorkBuddy.

If Tencent follows through on that rollout, WorkBuddy could receive a substantial capability upgrade at the model layer, allowing the product to take on more advanced productivity and agent-based tasks. More broadly, such a move would strengthen Tencent’s push to turn its AI research into practical tools for workplace and business use.

While the company has yet to make a formal launch announcement, the appearance of an internal testing interface strongly suggests that Hy4 is no longer a distant roadmap item. For Tencent, its release could mark the next major milestone in building a multi-model AI ecosystem that spans foundational research, enterprise productivity, and intelligent agent applications.