Tencent's Hy3 AI Model Sees 68x Usage Surge as Company Readies Hy4 Release

Enterprise Adoption Accelerates

Tencent's Hy3 language model has seen weekly usage climb more than 68 times compared to its predecessor model following the transition from preview to full production release. The jump signals growing enterprise and consumer adoption of Tencent's AI infrastructure as Chinese tech platforms race to embed generative models into core products.

The company has integrated Hy3 across multiple properties: WorkBuddy, its workplace collaboration suite; CodeBuddy, a developer-focused coding assistant; Yuanbao, its AI chatbot platform; and ima, Tencent's smart assistant. WorkBuddy alone recorded over 20 million PC visits in June, according to Tencent, though the company did not break out mobile traffic or active user counts.

The deployment comes as Tencent looks to monetize AI capabilities through existing distribution channels rather than launching standalone consumer AI products. By threading Hy3 into tools that already serve millions of enterprise customers and individual users, Tencent avoids the customer acquisition costs that have weighed on rivals focused on direct-to-consumer AI apps.

Hy4 in the Pipeline

Tencent announced plans to release Hy4, a larger-parameter successor model, in the near term. The company has not disclosed parameter count, training data scale, or a specific launch window. Industry observers expect Hy4 to exceed 100 billion parameters, aligning with the trajectory of competing models from Alibaba, Baidu, and ByteDance, though Tencent has historically prioritized inference efficiency over raw model size.

The decision to move quickly to a next-generation model reflects intensifying competition in China's crowded AI landscape. Alibaba's Qwen and Baidu's Ernie families have both released multiple iterations in the past year, each claiming performance gains on standard benchmarks. Tencent's rapid iteration suggests the company views model capability as a moving target that requires continuous investment to maintain competitive parity.

Chinese AI labs have also accelerated release cycles in response to improving open-source models and pressure from regulators to demonstrate domestic self-sufficiency in foundational AI technology. Tencent's timeline for Hy4, though unspecified, will likely be influenced by these external pressures as much as internal product roadmaps.

Revenue Context

Tencent reported second-quarter revenue of 204.79 billion yuan, providing financial context for its AI investments. The figure represents stable growth in Tencent's core gaming, social, and advertising businesses, which continue to fund the company's push into AI infrastructure and enterprise software.

Unlike startups that must justify AI spending to venture investors, Tencent can draw on cash flow from WeChat, Honor of Kings, and its advertising network to support long-cycle R&D. This financial cushion allows the company to pursue a strategy of embedding AI across existing products rather than chasing immediate monetization through subscription models or API sales.

The WorkBuddy traffic figure offers a glimpse into one monetization path. Enterprise collaboration tools represent a natural fit for generative AI features, with use cases spanning document drafting, meeting summaries, and workflow automation. If Tencent can convert WorkBuddy's 20 million monthly PC visitors into paying enterprise accounts, the AI investment begins to carry its own weight.

Competitive Landscape

Tencent's AI strategy sits within a broader battle among Chinese tech giants to control the next layer of enterprise software. Alibaba has embedded its Tongyi models into DingTalk, its workplace platform, while ByteDance has integrated Doubao into Feishu (known internationally as Lark). Each company is betting that AI capabilities will lock in enterprise customers and raise switching costs.

The 68-fold usage increase for Hy3 suggests Tencent is gaining traction, though the baseline remains opaque. Without disclosure of absolute usage numbers or the performance of the prior model, the percentage gain is difficult to contextualize. A 68x increase from a low base would be less meaningful than the same multiple applied to an already substantial user population.

What is clear is that Tencent views AI model development as table stakes rather than a differentiator. The company's emphasis on integration and distribution over standalone model performance reflects a pragmatic read of the market: users care less about parameter counts than about whether AI features solve immediate workflow problems. By threading Hy3 into tools people already use, Tencent lowers the friction for AI adoption and captures usage data that will inform Hy4's design.

The near-term release of Hy4 will test whether Tencent can maintain development velocity while scaling deployment. Larger models bring higher inference costs, which can erode margins if not offset by pricing power or efficiency gains. How Tencent balances model capability against operational cost will shape its ability to compete with rivals who are making similar bets on ever-larger language models.