Tencent's major AI model, Hy3, was officially launched on July 6, just about 10 weeks after its preview version was released.
Bernstein subsequently published a report, maintaining its "Outperform" rating on TENCENT, stating that while Hy3 is not a cutting-edge model, its performance is sufficient to support the deployment of Agent applications within the WeChat ecosystem, and market concerns over token costs and capital expenditures may be overblown.
Hy3 retains the architecture of 295 billion total parameters with 210 billion activated parameters, but shows significant optimization in key capabilities: tool calling stability has improved, the hallucination rate has dropped from 12.5% to 5.4%, multi-turn dialogue memory has been enhanced, and overall reasoning ability has reached a level comparable to GLM-5.1.
However, it still makes trade-offs in heavy programming and general reasoning, and the GQA architecture could be further iterated in the future through KV Cache compression technologies like MLA and DSA.
Bernstein points out that Hy3 is the first complete achievement of Tencent's new AI team, and its performance improvements validate that the model's pre-training, reinforcement learning, and evaluation systems are on the right track.
Public benchmark tests show that Hy3's overall performance is now significantly ahead of MiniMax's M3 model.
The next-generation model is expected to be launched as early as late 2026 or early 2027 and may introduce a brand-new pre-training architecture.
Analysts believe Tencent's AI strategy is shifting from model development to commercial implementation, with the core metric for measuring return on investment in the future likely to be the scale of Agent transactions, rather than chatbot usage.
The path for building Agent services around the WeChat ecosystem is becoming clearer, which is also key to Tencent's long-term AI value.
Agent Monetization More Critical Than Model Competition
Compared to the model's capabilities themselves, Bernstein is more focused on the monetization path for Tencent's Agent business.
The report notes that US internet platforms have seen a surge in Agent-driven e-commerce in recent years, which has since cooled down.
In contrast, Chinese internet platforms naturally possess closed-loop transaction capabilities, with a vast amount of user transaction behavior already existing within super-apps like WeChat, providing a more mature foundation for Agent monetization.
Bernstein expects that Tencent will not rush to charge ordinary users, and is more likely to adopt a B2B charging model.
Specifically, merchants subscribing to Agent services could gain more AI-driven traffic, richer AI tools, and deeper integration with the WeChat ecosystem; merchants who do not subscribe would correspondingly receive less traffic and resources.
This suggests Tencent's future AI revenue is more likely to come from merchant marketing and service fees, rather than direct consumer subscriptions.
Market May Be Overstating Token Cost Concerns
As global internet companies continue to increase their investment in AI infrastructure, Tencent's capital expenditure has similarly become a focus for investors.
Tencent has previously guided that capital expenditure will increase quarter-by-quarter through 2026, with corresponding depreciation and amortization expenses also rising.
However, Bernstein believes the market's current concerns about AI token consumption costs involve a degree of overreaction.
The report points out that a single ordinary chatbot conversation typically consumes only a few hundred tokens, whereas a genuine Agent transaction often requires tens of thousands of tokens.
This implies that token consumption for Tencent will only see an exponential increase when Agent transaction volume and gross merchandise value (GMV) grow substantially.
From a business model perspective, there may be a lag between user activity growth and revenue growth, but historical experience suggests that monetization is ultimately more a matter of timing rather than feasibility.
WeChat Agent Faces Internal Coordination Challenges
Nevertheless, Bernstein also notes that an important unresolved issue remains in Tencent's AI strategy.
The report states that while Tencent's reorganized AI team has made significant progress in model training infrastructure, it is understood that this team does not yet seem to have access to WeChat data; meanwhile, the upcoming WeChat AI Agent is primarily being developed independently by the WeChat team.
Analysts believe this organizational fragmentation will most likely be resolved through management coordination, but the reason why the two sides have not yet achieved deeper collaboration will also be a key point for investors to monitor regarding Tencent's AI strategy execution going forward.
Comments