Tencent's Hy3 launch strengthens AI Agent capabilities, and Morgan Stanley maintains Overweight with HK$650 target

Overview

This report is Morgan Stanley's company research/meeting-minutes on Tencent Holdings Ltd. (0700.HK), centered on Tencent Hy3's model capabilities, AI agent experience, product-ecosystem feedback loop, and valuation views after its open-source launch. The report argues that Hy3 has clearly strengthened capabilities in agent-style tasks and productivity applications and that real-world feedback from Tencent's internal product matrix can enhance model generalization and feed back into multiple products.

Core views

The core views are: first, Hy3's launch shows that Tencent's large model has continued to improve in agent capabilities and across multiple productivity tasks; second, Tencent Cloud API pricing shows relatively strong cost efficiency, supporting practical business adoption; third, Tencent's product ecosystem serves as both a validation field for model capabilities and a source of data and feedback for subsequent optimization; fourth, Morgan Stanley maintains an Overweight rating, an Attractive sector view, and a HK$650 target price for Tencent.

Analysis framework

The report combines company-event interpretation, model benchmark comparisons, product application cases, and a sum-of-the-parts valuation build-up. The AI section focuses on Hy3 architecture parameters, context length, SWE-Bench Pro score, Tencent Cloud API pricing, and deployment outcomes for WorkBuddy, Yuanbao Agent, IMA, Mavis Agent, WeRead, WeGame AI Game Assistant, and code generation. The valuation section uses a sum-of-the-parts approach, where the core business is valued through DCF and affiliate investments are handled at book value with a discount.

Methodology notes

• Valuation methodsSum of the Parts

  Base-case target price HK$650

  The target price is composed of HK$570 for core business and HK$80 per share for affiliate investments; core business uses DCF, while affiliate investments are based on reported book value and a 30% discount is applied to investment value.

• Valuation methodsDCF

  Core business valuation

  Core business valuation is HK$570, using a 10% discount rate and a 3% terminal growth rate.

• Model EvaluationSWE-Bench Pro

  Software engineering task benchmark

  Hy3 scored 57.9, above Hy3 Preview at 46 and DeepSeek V4 Pro at 55.4, but below Qwen 3.7 Max at 60.6 and GLM 5.2 at 62.1, to measure its agent capabilities on software development-related tasks.

• Rating FrameworkMorgan Stanley Relative Rating Framework

  Overweight and Attractive

  Overweight indicates that the stock's risk-adjusted total return over the next 12-18 months is expected to exceed the average for stocks in the analyst coverage universe; Attractive indicates that the industry view is compelling relative to a relevant broad-based benchmark over the next 12-18 months.

Asset mapping & comparison

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

• Tencent Holdings Ltd.（0700.HK）

  The report target company and main beneficiary of Hy3 and Tencent's AI ecosystem

  Strengths

  It has a broad product matrix that can provide real-world feedback for Hy3 and create a virtuous cycle in which model and product experience reinforce each other; games, advertising, social networking, and AI applications form a diversified business base.

  Weaknesses

  The report does not quantify Hy3's direct contribution to revenue and profit in detail, and AI commercialization pace still needs further validation.

  Comparison

  Hy3 is above Hy3 Preview and DeepSeek V4 Pro on SWE-Bench Pro, but below Qwen 3.7 Max and GLM 5.2.

  Risks

  Uncertainty in game regulation, intensified competition for social network and advertising budgets, stricter antitrust scrutiny, and U.S.-China tensions.

• Tencent Cloud

  Entry point for Hy3 API commercialization and enterprise customer calls

  Strengths

  Relatively cost-efficient API pricing, with lower cache-hit input pricing, which supports higher practical adoption.

  Weaknesses

  Cloud model services still face competition on pricing, performance, and ecosystem from peers.

  Comparison

  The report does not provide systematic price comparisons of Tencent Cloud API versus other cloud providers, only emphasizes its own pricing.

  Risks

  Inference cost, pricing competition, adoption speed, and reliability requirements from customers.

• WorkBuddy

  Product application case of improved Hy3 capabilities

  Strengths

  Task completion improved from 72% to 90%, and average completion time fell by 34%, indicating that stronger agent capabilities can translate into efficiency gains.

  Weaknesses

  The report does not disclose sample size, evaluation methodology, or long-term retention effects.

  Comparison

  As an internal use case, it is mainly used to show pre- and post-upgrade improvement.

  Risks

  The complexity of real user scenarios, data definitions, and the sustainability of improvements still need observation.

• Yuanbao Agent

  Application case of Hy3 in office and life-service agent scenarios

  Strengths

  Outperformed GLM 5.1 and several domestic models in comprehensive office and life-service agent scenario evaluations.

  Weaknesses

  The report does not disclose full leaderboards, weighting, or detailed sample data.

  Comparison

  Performing stronger than GLM 5.1 and multiple domestic models.

  Risks

  Rapid iteration by competing models may quickly narrow the advantage.

Key data

• Stock RatingOverweightThe table shows stock rating as Overweight.
• Industry ViewAttractiveThe table shows industry view as Attractive.
• Target PriceHK$650.00Base-case target price based on sum-of-the-parts.
• CloseHK$452.00Closing price as of July 6, 2026.
• Upside to Price Target44%The report table shows up/downside to price target as 44%.
• Hy3 Parameter Size295B parameters, 21B activated parametersHy3 continues the MoE architecture from Hy3 Preview.
• Context Length256k-tokenThe report states Hy3 supports 256k-token context.
• Tencent Cloud API PricingInput/output CNY 1.0/4.0 per million tokens; cache-hit input CNY 0.25 per million tokensA practical path reflecting cost efficiency.
• SWE-Bench Pro Score57.9Above Hy3 Preview 46 and DeepSeek V4 Pro 55.4, below Qwen 3.7 Max 60.6 and GLM 5.2 62.1.
• WorkBuddy Task Completion Rate72% to 90%Average completion time shortened by 34%.
• 2026e RevenueRMB 826.9 billionRevenue, net 12/26e in the financial table.
• 2026e ModelWare Net ProfitRMB 240.2 billionModelWare net inc 12/26e in the financial table.
• 2026e P/E15.1x12/26e P/E in the financial table.

Impact & implications

If Hy3's capability improvements continue to translate into better user experience, task completion rates, and commercialization efficiency in Tencent's internal products, AI spending may become a source of strength rather than just a cost item, enhancing competitiveness in Tencent Cloud, office and life-service agents, content products, gaming, and advertising. The report maintains a constructive rating on the stock, reflecting that improved AI capability, game execution, advertising share, and 2C AI adoption ramp-up may become key upside drivers over the next 12-18 months.

Risks

• Uncertainty around game regulation.
• Increased competition for social network and advertising budgets from emerging entertainment formats.
• Increased antitrust scrutiny and uncertainty from U.S.-China tensions.
• AI adoption speed and commercialization realization may be below expectations.
• If AI capability improvements fail to consistently translate into user growth, retention, or monetization, the investment case could be affected.

What to watch

• Call volume, customer adoption, and pricing competition for Hy3 on Tencent Cloud API.
• AI product retention, conversion, and commercialization metrics for WorkBuddy, Yuanbao Agent, IMA, Mavis Agent, WeRead, and WeGame AI Game Assistant.
• Relative ranking changes in agent capability benchmarks such as SWE-Bench Pro for future versions of Hy3.
• Launch performance of Tencent's new games in domestic and overseas markets.
• Changes in social and short-video advertising market share.
• Whether 2C AI adoption ramp-up can drive better user experience and business growth.