Morgan Stanley is positive on Tencent's improvements in Hy3 agent capabilities and its positive product ecosystem cycle
AI summary card
Morgan Stanley is positive on Tencent's improvements in Hy3 agent capabilities and its positive product ecosystem cycle
The report believes Tencent's Hy3 has made notable progress in software development, office productivity, financial modeling, frontend design, and game production tasks. Combined with real-world feedback from Tencent's ecosystem, this could strengthen the deployment capabilities of AI products.
- Tencent open-sourced Hy3 on July 6, retaining the MoE architecture of Hy3 Preview, with 295B parameters, 21B active parameters, and support for a 256k-token context.
- Hy3 scored 57.9 on SWE-Bench Pro, above Hy3 Preview's 46 and DeepSeek V4 Pro's 55.4, but below Qwen 3.7 Max's 60.6 and GLM 5.2's 62.1.
- The WorkBuddy task resolution rate increased from 72% to 90%, while average completion time decreased by 34%, indicating significant improvement in enterprise productivity scenarios.
- The report gives Tencent a target price of HK$650.00, implying 44% upside from the July 6 closing price of HK$452.00. The current rating is Overweight.
Report interpretation
Overview
This report is a Morgan Stanley company research conference note on Tencent Holdings Ltd., focusing primarily on Tencent's progress in agent capabilities and product experience following the open-sourcing of Hy3. The report believes Hy3 has achieved substantial improvements across multiple productivity tasks and is receiving extensive, real-world application feedback through Tencent's own product matrix, thereby driving continuous optimization of model capabilities and product experience.
Core views
The core view is that Hy3's capability improvements are expanding from model benchmark testing to practical applications across Tencent's product ecosystem, including WorkBuddy, Yuanbao Agent, IMA, Mavis Agent, WeRead, WeGame AI Game Assistant, and code generation. Tencent Cloud's API pricing is cost-efficient, and improvements in model capabilities should enhance the competitiveness of Tencent's AI products while supporting the company's long-term value in 2C AI adoption, gaming, advertising, and social entertainment businesses.
Analysis framework
The report combines event commentary with product progress tracking, analyzing Hy3's model architecture, API pricing, benchmark performance, and the effectiveness of deployment across Tencent's ecosystem products. The valuation uses a sum-of-the-parts approach, with core businesses valued using DCF and associates valued at book value with a discount applied.
Methodology notes
The target price consists of core business value and associate investment value
The report gives a base-case target price of HK$650, comprising HK$570 for core businesses and HK$80 per share for associate investments. Core businesses are valued using a DCF with a 10% discount rate and 3% perpetual growth rate, while associate investments are based on book value with a 30% discount applied to investment value.
Measures model capabilities using software engineering task benchmarks
Hy3 scored 57.9 on SWE-Bench Pro, above Hy3 Preview and DeepSeek V4 Pro but below Qwen 3.7 Max and GLM 5.2, indicating significant progress in software development and agent capabilities compared with the prior generation.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- Tencent Holdings Ltd. (00700.HK)Research subject
- Strengths
- Hy3 model capabilities are improving, Tencent Cloud pricing is cost-efficient, and the product matrix can provide real-world feedback to create a positive cycle of AI capability iteration. The rating is Overweight, with the target price implying 44% upside.
- Weaknesses
- The AI progress disclosed in the report remains primarily focused on model releases, benchmark testing, and improvements in product scenarios. The long-term contribution of commercialized revenue still requires further validation.
- Comparison
- Hy3's SWE-Bench Pro score is higher than those of Hy3 Preview and DeepSeek V4 Pro, but lower than those of Qwen 3.7 Max and GLM 5.2.
- Risks
- Regulatory uncertainty in the gaming industry, intensifying competition for social-network user time and advertising budgets, tightening antitrust regulation, and heightened US-China tensions.
Key data
- Hy3 architecture295B parameters, 21B active parameters, 256k-token contextTencent open-sourced Hy3 on July 6, retaining the MoE architecture of Hy3 Preview.
- Tencent Cloud API pricingRMB 1.0/4.0 per million input/output tokens, respectively; cached input at RMB 0.25 per million tokensThe report describes this as a cost-efficient and practical approach.
- SWE-Bench Pro57.9Above Hy3 Preview's 46 and DeepSeek V4 Pro's 55.4, but below Qwen 3.7 Max's 60.6 and GLM 5.2's 62.1.
- WorkBuddy task resolution rateIncreased from 72% to 90%Average completion time decreased by 34%.
- Target priceHK$650.00Base-case sum-of-the-parts valuation.
- Current share priceHK$452.00Closing price as of July 6, 2026.
- Target price upside44%Based on the target price of HK$650.00 and closing price of HK$452.00.
- 2026e revenueRMB 826.9 billionEstimate based on the Morgan Stanley ModelWare framework.
- 2026e EPSRMB 25.88Based on consensus methodology.
Impact & implications
From an investment perspective, Hy3's capability improvements and the product feedback mechanism within Tencent's ecosystem could strengthen Tencent's competitive advantages in AI applications, cloud services, office productivity, and game content production. Faster 2C AI adoption, successful execution of new games, and increased advertising share could support valuation upside; however, regulation, competition, and US-China relations remain key uncertainties.
Risks
- Regulatory uncertainty in the gaming industry could affect the release and commercialization timelines of new games.
- Emerging entertainment formats could intensify competition for social-network user time and advertising budgets.
- Tightening antitrust regulation and heightened US-China tensions could affect the company's valuation and operating environment.
- The pace of AI product adoption, user retention, and commercialization results still require continuous validation.
What to watch
- The continued deployment performance of Hy3 across products including WorkBuddy, Yuanbao Agent, IMA, Mavis Agent, WeRead, and WeGame AI Game Assistant.
- Whether 2C AI adoption accelerates and translates into user growth, improved retention, or increased monetization.
- The impact of Tencent Cloud's AI API pricing strategy on developer ecosystem growth and enterprise customer adoption.
- The execution of new game launches domestically and internationally, as well as changes in advertising market share.
- Whether subsequent ratings, target prices, and 2026-2028 revenue, EPS, and EBITDA forecasts are revised upward.