Bernstein comments on Tencent’s Hy3 model: AI capabilities are steadily improving, and the commercialization path for agents is becoming increasingly clear.

```

The official version of Tencent’s Hy3 AI large model was released on July 6, just about 10 weeks after the preview version went online. Bernstein immediately issued a report, maintaining an “outperform” rating for Tencent, stating that although Hy3 is not a cutting-edge model, its performance is sufficient to support Agent applications within the WeChat ecosystem. Concerns about Token costs and capital expenditure in the market may be exaggerated.

Hy3 continues with a structure of 295 billion total parameters and 21 billion active parameters, but has significantly optimized key capabilities: improved tool call stability, reduced hallucination rate from 12.5% to 5.4%, enhanced multi-turn conversation memory, and overall reasoning ability reaching GLM-5.1 level. However, it still makes trade-offs in heavy programming and general reasoning, and the GQA architecture can be further iterated in the future via KV Cache compression technologies such as MLA and DSA.

Bernstein points out that Hy3 is the first complete achievement of Tencent’s new AI team, and its performance improvements prove that model pre-training, reinforcement learning and evaluation systems are already on track. Public benchmarking shows that Hy3’s overall performance is clearly ahead of MiniMax M3. The next-generation model is expected to be launched as early as the end of 2026 or beginning of 2027, and may introduce brand-new pre-training architecture.

Analysts believe, Tencent’s AI strategy is shifting from model R&D to commercial implementation, and in the future, the core metric for measuring return on investment will be Agent transaction scale, rather than chatbot usage. The service pathway for Agents built around the WeChat ecosystem is gradually becoming clearer, which is also the key to Tencent AI’s long-term value.

Agent commercialization is more important than model competition

Compared to the model’s capabilities, Bernstein is more concerned with the commercialization path of Tencent’s Agent business.

The report notes that U.S. internet platforms have seen Agent e-commerce booms in recent years, but quickly cooled; by contrast, Chinese internet platforms naturally have a closed transaction loop, and a large volume of user transactions already exist within super apps like WeChat, making the foundation for Agent commercialization much more mature.

Bernstein expects that, Tencent will not rush to charge ordinary users, but is more likely to adopt a B-end charging model. Specifically, merchants who subscribe to Agent services can gain more AI traffic, richer AI tools, and deeper connections to the WeChat ecosystem, while unsubscribed merchants will correspondingly receive less traffic and resources. This means that Tencent’s future AI revenue is more likely to come from merchant marketing and service fees rather than direct consumer subscriptions.

Concerns about Token costs may be exaggerated by the market

As global internet companies continue to increase investment in AI infrastructure, Tencent’s capital expenditure has also become a focus for investors.

Tencent has previously indicated that capital expenditure will increase quarter by quarter in 2026, and corresponding depreciation and amortization expenses will also rise. However, Bernstein believes that the market’s current concerns about AI Token consumption costs are somewhat overstated. The report points out that a regular chatbot conversation usually only consumes hundreds of Tokens, while real Agent transactions often require tens of thousands of Tokens.

This means, only when the Agent transaction volume and total transaction amount (GMV) grow on a large scale will Tencent’s Token consumption see an exponential increase. From a business model perspective, there may be a time lag between user activity growth and revenue growth, but historical experience indicates that commercialization is ultimately more a matter of time than a question of feasibility.

WeChat Agent still faces internal coordination challenges

However, Bernstein also points out that there is still an important unresolved issue in Tencent’s AI strategy.

The report says that Tencent’s restructured AI team has made obvious progress in model training infrastructure, but as understood, the team does not seem to have obtained data usage rights for WeChat. Meanwhile, the upcoming WeChat AI Agent is mainly independently developed by the WeChat team.

Analysts believe, the segmentation caused by this organizational structure will very likely be resolved through management coordination in the end, but why both sides have not yet achieved deeper collaboration will also become an important observation point for investors monitoring the execution of Tencent’s AI strategy in the future.

Risk warnings and disclaimerThe market entails risk and investments should be made cautiously. This article does not constitute personal investment advice and has not considered the individual investment objectives, financial situation or needs of users. Users should consider whether any opinions, views or conclusions in this article are suitable for their specific circumstances. Investments based on this are at one’s own risk. ```