Bernstein is positive on Tencent's Hy3 progress and AI agent monetization potential in the WeChat ecosystem

Overview

This report is a Bernstein company research brief on Tencent Holdings Ltd, focusing on the investment implications after the official launch of Tencent's Hy3 model. It argues that the Hy3 formal release has made substantial improvements in key performance metrics compared with the preview version; although it is still not a global frontier model, it is already capable of supporting Tencent in advancing agent services within its own ecosystem. The report maintains Tencent at Outperform with a target price of HK$780.

Core views

Core views include: first, the Hy3 formal release demonstrates that Tencent's new AI team is pursuing an effective path in pretraining, reinforcement learning, and evaluation framework; second, market concerns about Tencent's AI token costs are overly strict, because high token burn is dependent on a real takeoff in agent transaction volume and GMV; third, Chinese internet platforms have a stronger natural transaction-intent base than U.S. platforms, and Tencent's AI commercialization is more likely to monetize through charging merchants for agent traffic, AI tools, and WeChat ecosystem connectivity; fourth, coordination of data and organization between Tencent's AI team and WeChat team remains a key management issue.

Analysis framework

The report combines model benchmark tests, Tencent ecosystem scenarios, capital expenditure and token burn logic, and consumer AI commercialization pathways in its assessment. For valuation, it uses a 20x FY+1 P/E multiple, corresponding to an HK$780 per-share target price.

Methodology notes

• Valuation methodologyFY+1 P/E valuation

  20x FY+1 PE multiple

  The report uses the forecast earnings of fiscal quarters 5 through 8 as an FY+1 basis and applies a 20x multiple, valuing Tencent at HK$780 per share.

• AI model evaluationModel capability benchmark comparison

  Tool calling stability, hallucination control, multi-turn retention, inference capability

  The report cites Tencent's disclosed Hy3 benchmark results and states that tool-calling stability, hallucination rate, and MRCR multi-turn retention have improved significantly versus the preview version, approaching GLM-5.1-level capabilities.

• Commercialization frameworkAgentic AI transactions and merchant monetization

  agentic AI traffic monetization

  The report believes agentic AI requires substantial tokens, but cost pressure becomes meaningful only after transaction volume and GMV scale up significantly; monetization is more likely to come from merchants paying for agent traffic, AI tools, and WeChat ecosystem connection.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Tencent Holdings Ltd / 00700.HK

  Primary covered name; Hy3 model progress and WeChat ecosystem commercialization are the key investment themes.

  Strengths

  Tencent has a WeChat ecosystem, transaction intent, merchant base, and multiple business scenarios; the Hy3 formal version shows progress in the AI team's model training and evaluation framework.

  Weaknesses

  Hy3 is still not a frontier model, and heavy coding plus general reasoning are not the primary focus; data and organizational alignment between the AI team and WeChat team are not yet fully connected.

  Comparison

  The report states that Hy3 reaches a GLM-5.1 level for the capabilities Tencent needs and clearly outperforms Minimaxi M3; the transaction-intent base of Chinese platforms is also stronger than that of U.S. platforms, which must re-learn click-driving behavior for consumption.

  Risks

  Rising capital expenditure, delayed token-cost pass-through, constrained access to WeChat data, competition in gaming and advertising, and macro consumption and regulatory risks.

Key data

• RatingOutperformBernstein's rating on Tencent Holdings Ltd.
• Target priceHKD 780.00The target price given in the report.
• Closing priceHKD 452.00Closing date is 2026-07-06.
• Implied upside73%Based on HKD 452.00 closing price versus HKD 780.00 target price.
• Hy3 model size295bn parameters, 21bn active parametersThe formal and preview versions have the same parameter scale.
• Hallucination control12.5% down to 5.4%Tencent-reported Hy3 performance improvement data cited by the report.
• MRCR multi-turn retention42.9% vs. 75.1%The report identifies this as one of the key performance metrics.
• F26E revenueCNY 828,757mFrom the report's financial forecast table.
• F27E adjusted EPSCNY 34.91Adjusted EPS forecast in the report.
• F26E adjusted P/E13.1xFrom the report's valuation metrics table.

Impact & implications

The investment implication is that the Hy3 formal release has boosted market confidence in Tencent's AI execution capability and reduced the severity of the negative narrative that Tencent is lagging in large-model competition and facing uncontrolled costs. If WeChat agents are gradually rolled out and a merchant-side monetization model takes hold, Tencent could convert existing social traffic, transaction intent, and merchant ecosystem into a new AI commercialization channel. That said, capital expenditure, depreciation and amortization, organizational coordination, and regulatory factors will still affect the timing of earnings expansion.

Risks

• Macro risks, including tighter credit conditions and weak retail consumption.
• Fluctuations in user engagement on Tencent's platforms.
• Competition from other internet platforms for gaming and advertising demand.
• Chinese antitrust and related regulatory risks.
• Depreciation and amortization pressure from rising AI capital expenditure.
• If agent transaction volume and GMV expand quickly, token consumption and inference costs may rise materially.
• If AI-model and WeChat-team data and organizational alignment is not resolved, agent product deployment could be affected.

What to watch

• Whether a next-generation Hy3 model will be released by late 2026 or early 2027, and whether it includes new pretraining.
• The release cadence, feature scope, and user engagement of WeChat AI agents.
• Whether merchants are willing to pay for agent AI traffic, AI tools, and WeChat ecosystem connectivity.
• Quarter-over-quarter changes in Tencent's capital expenditure and the impact on depreciation and amortization.
• The actual relationship between agent transaction volume, GMV, and token consumption.
• Whether management resolves data-access and collaboration issues between the AI model team and the WeChat team.