Tencent open-sources Hy3 Preview, reinforcing the AI model and ecosystem deployment narrative

Overview

This report focuses on Tencent Holdings' next-generation large language model, Hy3 Preview, released and open-sourced on April 23, 2026. The report positions it as the first step in Tencent's rebuilding of the HY foundation model and highlights the model's progress in architecture, long context, inference efficiency, code generation, and Agent capabilities, as well as its early application results in Tencent products such as Yuanbao, CodeBuddy, WorkBuddy, QQ, and ima.

Core views

The core view is that Hy3 Preview enhances Tencent's competitiveness in AI infrastructure and application ecosystem, and that its open-source strategy will help absorb real developer and user feedback, laying the groundwork for iteration toward the formal release. Tencent Cloud is advancing commercialization through competitive API pricing and customized token plans, with personal-tier pricing starting from Rmb28/month. On the stock side, Morgan Stanley maintains its Overweight rating on Tencent, with an Attractive industry view and a target price of HK$650.

Analysis framework

The report analyzes the topic through a combination of company event tracking, model capability benchmarking, observation of product ecosystem deployment, and valuation framework analysis. On the technical side, it focuses on model parameter scale, activated parameters, context length, inference efficiency, benchmark scores, and real product applications; on the investment side, it is based on a sum-of-the-parts valuation, combined with DCF for the core business and assessment of associate investment value to derive the target price.

Methodology notes

• Valuation methodsSum-of-the-Parts Valuation

  The HK$650 target price consists of HK$560 for the core business and HK$90/share for associate investments.

  The report uses sum of the parts as the base-case valuation method, valuing Tencent's core business and associate investments separately and then adding them together.

• Valuation methodsDCF

  The core business value of HK$560 is based on DCF valuation.

  The DCF assumptions include a 10% discount rate and a 3% terminal growth rate, used to estimate the value of the core business.

• Valuation methodsInvestment Value Discount

  Associate investments at HK$90/share are based on reported book value with a 30% discount applied to investment value.

  This treatment reflects a discount for monetization potential or valuation uncertainty of associate investments.

• Financial ModelMorgan Stanley ModelWare

  Unless otherwise stated, metrics are based on the Morgan Stanley ModelWare framework.

  Metrics in the report such as forecast EPS, revenue, EBITDA, net profit, valuation multiples, and ROE are mainly derived from Morgan Stanley Research estimates or consensus data.

Asset mapping & comparison

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

• Tencent Holdings Ltd. (0700.HK)

  Company covered by the report

  Strengths

  Hy3 Preview open-sourcing, improved inference efficiency, deployment across multiple products within the Tencent ecosystem, Overweight rating, and upside potential from the current price to the target price.

  Weaknesses

  Benchmark performance still trails some competing products such as GLM-5, Kimi-K2.5, and Claude-Opus-4.6; AI commercialization still requires further validation.

  Comparison

  Its SME-Bench Verified score is 74.4%, close to GLM 4.7's 73.8%, but below Claude-Opus-4.6's 80.8%.

  Risks

  Game regulation, competition in social and advertising, antitrust regulation, and uncertainty in China-US relations.

• Tencent Cloud

  Commercialization and API platform for Hy3 Preview

  Strengths

  Competitive API pricing, customized token plans, and personal-tier pricing starting from Rmb28/month.

  Weaknesses

  Price competition may compress unit economics, and demand conversion still needs to be observed.

  Comparison

  The report does not provide a full peer pricing comparison for cloud services, only emphasizing pricing competitiveness.

  Risks

  Large-model inference costs, customer adoption speed, and competition from peer models.

• Yuanbao、CodeBuddy、WorkBuddy、QQ、ima

  Early applications and product ecosystem validation scenarios

  Strengths

  The report states that early applications have delivered meaningful real-world performance improvements.

  Weaknesses

  The report does not quantify specific user growth, retention, or revenue contribution for each product.

  Comparison

  These products reflect Tencent's advantage in internal ecosystem integration rather than a direct comparison of standalone external model capabilities.

  Risks

  Whether product experience improvements can translate into user scale, paid conversion, or advertising efficiency remains to be verified.

Key data

• Model release date2026-04-23Tencent released and open-sourced Hy3 Preview.
• Model architectureMoE, 295B parameters, 21B activated parameters, 256k token contextIt supports both fast and slow inference and is positioned as the first step in rebuilding the HY foundation model.
• Inference efficiency improvement40%Tencent's full-stack optimization brings improved inference efficiency and lower overall costs.
• SME-Bench Verified score74.4%Higher than Hy2's 53%, close to GLM 4.7's 73.8%, and below GLM-5's 77.8%, Kimi-K2.5's 76.8%, and Claude-Opus-4.6's 80.8%.
• Tencent Cloud personal-tier priceRmb28/月起Hy3 Preview offers competitive API pricing and customized token plans.
• Stock ratingOverweightMorgan Stanley's rating on Tencent Holdings Ltd.
• Industry viewAttractiveThe covered industry is China Internet and Other Services.
• Target priceHK$650.00The base case uses a sum-of-the-parts valuation.
• Current share priceHK$495.20Closing price on April 23, 2026.
• Upside to target price31%Relative to the HK$495.20 closing price.
• 2026e revenueRmb828.0bnMorgan Stanley forecast.
• 2026e EBITDARmb344.1bnMorgan Stanley forecast.
• 2026e P/E17.1xDisclosed in the valuation table.

Impact & implications

The release of Hy3 Preview strengthens Tencent's narrative in large-model infrastructure, developer ecosystem, and AI enablement across internal products. If open-source feedback and subsequent formal-version iterations proceed smoothly, Tencent may enhance model use cases and commercialization efficiency through cloud services, coding tools, office products, and social products. For the stock, improved AI adoption is one of the upside risks, but the investment conclusion still needs to be considered alongside games, advertising, social network competition, and the regulatory environment.

Risks

• Regulatory uncertainty in the gaming industry.
• Social networks and advertising budgets face intensifying competition from emerging forms of entertainment.
• Tighter antitrust regulation and heightened China-US tensions.
• AI adoption and commercialization progress may fall short of expectations.
• Although model capabilities have improved significantly, some benchmark tests still lag leading competitors.
• Morgan Stanley discloses that it has or may have investment banking and other service relationships with Tencent and many covered companies; investors should pay attention to potential conflicts of interest.

What to watch

• Developer feedback after Hy3 Preview is open-sourced, the pace of model iteration, and the release timing of the formal version.
• Whether expanded training scale and reinforcement learning investment continue to improve the model's reasoning, code generation, and Agent capabilities.
• Usage rates and commercialization outcomes of AI features in products such as Yuanbao, CodeBuddy, WorkBuddy, QQ, and ima.
• Tencent Cloud's Hy3 Preview API pricing, token plans, and enterprise customer adoption.
• Performance of new game launches and domestic and overseas game gross receipts.
• Changes in market share for social and short-video advertising.
• Changes in regulatory policy, antitrust enforcement, and China-US relations.