Strong momentum in China software AI product launches, with Agentic AI driving token consumption and commercialization expansion

Overview

This report is Goldman Sachs' product tracker on new AI product launches in China's software sector since the beginning of 2026. The report notes that momentum in launching AI Agents and new AI features in software applications remains strong, driven by improvements in training and inference efficiency, lower token costs, stronger open-source model capabilities, greater adoption by consumer and enterprise users, and clearer commercialization paths.

Core views

The core views are: first, enhanced Chinese foundation models are enabling more complex agentic tasks such as work assistants, call centers, industry experts, and coding scenarios, while bringing an inflection point in token economics; second, AI-native applications are expanding across productivity, entertainment, design, documents, video, short dramas, and other scenarios, with MAU likely to continue growing; third, as tokens become cheaper and usage scenarios broaden, positive ROI for end users will further drive adoption and payment, and software vendors may gradually shift from subscription fees to pricing based on token usage volume.

Analysis framework

The report reviews software product launches from the beginning of 2026 to the report date, groups them into AI model, AI agent, and AI software applications, compares companies, products, key upgrades, and prior-generation products, and combines case studies of key companies to analyze commercialization, user adoption, and product feature expansion.

Methodology notes

• product_trackerAI Product Launch Tracker

  Observe industry momentum through product launches and feature upgrades

  The report uses new models, new Agents, and new AI software applications as core evidence, comparing feature upgrades, use cases, prior-generation products, and commercialization signals to assess the expansion trend of AI applications in China's software sector.

• investment_frameworkGS Factor Profile

  Comparison of growth, financial returns, valuation multiples, and composite percentile rankings

  Goldman Sachs states that its Factor Profile compares a stock's growth, financial returns, valuation multiples, and composite metrics against the market and industry peers to provide investment context for individual names.

• investment_frameworkM&A Rank

  Probability ranking of becoming an M&A target

  Goldman Sachs states that its M&A framework classifies covered companies into ranks 1 to 3 by probability of being acquired, where 1 indicates high probability, 2 indicates medium probability, and 3 indicates low probability.

Asset mapping & comparison

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

• SenseTime / Sensetime / 0020.HK

  Beneficiary of AI models and AI office applications; Goldman Sachs rating Buy

  Strengths

  Launched SenseNova 6.7 Flash-Lite, SenseNova U1, SenseNova-Skills, and AI Office Raccoon, covering scenarios such as understanding and generation, desktop Agents, infographic generation, PPT generation, data analysis, and deep research.

  Weaknesses

  Commercialization still depends on user adoption, enterprise integration, and realization of token economics, and the report does not provide specific revenue contribution estimates.

  Comparison

  Compared with traditional office software, SenseTime places greater emphasis on end-to-end automation, an open ecosystem, and multimodal Agent capabilities.

  Risks

  Intensifying model competition, long enterprise customer deployment cycles, and lower-than-expected usage frequency of AI features.

• Meitu / 1357.HK

  Beneficiary of AI video, design, and visual content generation; Goldman Sachs rating Buy

  Strengths

  Expansion of products such as Wink, Kaipai, RoboNeo, MVLAND, Artflo, Picchi, and MeituHub, with MVLAND rapidly generating music visual content through multi-Agent collaboration; management emphasized rising ARPU spending and strong user demand.

  Weaknesses

  User retention, paid conversion, and content quality stability for entertainment and creative tools still require continued validation.

  Comparison

  Compared with office-oriented Agents, Meitu is more focused on visual creation, music videos, AI short dramas, and design commercialization scenarios.

  Risks

  Intense competition in generative content, fluctuations in users' willingness to pay, and regulatory and copyright risks.

• Hundsun

  Buy recommendation in financial software

  Strengths

  The report lists it as a Buy recommendation in financial software, benefiting from AI adoption in financial industry software and upgrades to enterprise applications.

  Weaknesses

  The main text does not provide detailed product case studies, and the investment thesis mainly comes from the recommendation list rather than detailed analysis in this report.

  Comparison

  Compared with AI content generation companies, Hundsun is more focused on vertical financial software applications.

  Risks

  Uncertainty around financial IT budgeting cycles, regulatory changes, and product rollout pace.

• Tuya

  Buy recommendation in IoT software

  Strengths

  The report lists it as a Buy recommendation in IoT software, potentially benefiting from the combination of smart devices and AI applications.

  Weaknesses

  The main text does not provide detailed product case studies, and the evidence mainly comes from the recommendation list.

  Comparison

  Unlike office and content tools, Tuya is more focused on IoT software and device ecosystems.

  Risks

  Uncertainty around IoT demand cycles, customer budgets, and actual monetization of AI features.

• Glodon / 002410.SZ

  Construction AI Agent name; Goldman Sachs rating Sell

  Strengths

  Owns data assets in the construction industry, develops core AI products such as PMLead and PMSmart, and has established an AI engineering department; product upgrades are underway in construction cost and construction management scenarios.

  Weaknesses

  Despite its AI Agent positioning, Goldman Sachs still assigns a Sell rating, indicating that its investment appeal may be constrained by valuation, growth, or fundamentals.

  Comparison

  Glodon is more focused on vertical construction-industry Agents rather than general office or entertainment AI applications.

  Risks

  Construction industry demand, pace of customer expansion, actual ROI of AI products, and valuation digestion risk.

Key data

• Coverage windowBeginning of 2026 to 2026-07-17The report states that it reviewed new software product launches since the beginning of 2026.
• SenseNova 6.7 Flash-Lite token consumptionAbout 60% lower than pure-text AgentsThe table shows that SenseTime's lightweight native multimodal Agent model can reduce token consumption.
• Spark X2 parameter scale293B MoE LLMiFlytek Spark X2 is based on domestic Ascend computing power, with inference capability improving by about 50% versus the previous generation and supporting more than 130 languages.
• Share of calls to MiracleVision V6 in Meitu applications96.3%The table states that the sixth-generation visual foundation model MiracleVision V6 covers 96.3% of Gen-AI calls in Meitu applications.
• Goldman Sachs global stock coverage3,104 stocksAs of 2026-07-01, Goldman Sachs Global Investment Research had investment ratings on 3,104 stocks.
• Goldman Sachs global rating distributionBuy 50%, Hold 34%, Sell 16%From the disclosure table.

Impact & implications

For investment implications, the report reinforces the view that commercialization of AI applications in China's software sector is moving from proof of concept toward more quantifiable revenue models. Falling token costs improve end-user ROI, while Agentic AI expands use cases and call frequency, potentially driving token consumption, MAU growth, and enterprise customer adoption; however, differences remain across companies in product conversion capability, scenario value, and ability to absorb valuations.

Risks

• Actual user adoption, retention, and paid conversion for AI Agents and AI-native applications may fall short of expectations.
• Declining token costs may fail to fully translate into positive end-user ROI or revenue growth for software vendors.
• Intensifying industry competition may lead to pricing pressure, feature homogenization, and rising customer acquisition costs.
• Enterprise deployment, data integration, and security/compliance requirements may slow commercialization.
• Annual and quarterly subscriptions remain the core revenue source, and the speed of transition toward usage-based token fees is uncertain.
• Generative content, data usage, copyright, and cross-border regulation may affect the expansion of AI applications.

What to watch

• Installation volume, active users, and enterprise integration progress of SenseTime AI Office Raccoon, SenseNova-Skills, and SenseNova U1.
• ARPU, MAU, paid conversion, and content generation quality of Meitu MVLAND, Artflo, Kaipai, and RoboNeo.
• Real ROI of enterprise Agents in ERP, HR, CRM, cybersecurity, construction management, and invoice/tax scenarios.
• The impact of token pricing, model inference efficiency, and open-source model capability changes on software business models.
• Changes in the revenue mix between subscription fees and usage-based token fees for software vendors.
• Whether Goldman Sachs subsequently adjusts ratings, price targets, or earnings forecasts for SenseTime, Meitu, Hundsun, Tuya, and Glodon.