目前，通义DeepResearch的模型、框架和方案均已全面开源，用户可在Github、Hugging Face和魔搭社 区下载模型和代码。 模型训练 通义团队构建了一套以合成数据驱动、贯穿预训练与后训练的完整训练链路。模型能力基于一种多阶段 数据策略，旨在创建海量高质量的代理训练数据，而无需依赖昂贵的人工标注。 同时，该链路以Qwen3-30B-A3B模型为基座进行优化，为了确保训练的高效性与模型的鲁棒性，团队 创新性地设计了覆盖真实环境与虚拟环境的RL算法验证与真实训练模块，并结合高效异步强化学习算 法及自动化数据策展（Data Curation）流程，显著提升了模型的迭代速度和泛化能力。 在推理阶段，团队设计了ReAct和基于自研的IterResearch的Heavy两种模式，前者用于精准考察模型的 基础内在能力，后者则通过test-time scaling策略，充分挖掘并展现了模型所能达到的性能上限，即使在 长任务中，也能实现高质量的推理。 模型性能 在Humanity's Last Exam（HLE）、BrowseComp、BrowseComp-ZH、GAIA、xbench-DeepSearch、 ...

Core Viewpoint - The investment narrative for AI infrastructure in China is being reignited due to the explosive demand for enterprise-level applications and the accelerated adoption of large models by companies [1][2]. Group 1: AI Infrastructure and Market Dynamics - Goldman Sachs reports that the daily token consumption of enterprise-level large models in China reached 10.2 trillion in the first half of 2025, a staggering increase of 363% compared to the second half of 2024 [2][4]. - Alibaba, with a 47% market share in China's public cloud sector, is well-positioned for growth due to its leading model capabilities and diversified chip supply [3][7]. - The capital expenditure of Chinese cloud service providers (CSPs) is expected to grow by 39% year-on-year in Q3 2025, providing a hardware foundation for sustained AI cloud revenue growth [6][7]. Group 2: Competitive Landscape - Among model providers, Alibaba, ByteDance, and DeepSeek are the top choices for Chinese enterprises deploying general large models [5]. - Alibaba has penetrated most of the Fortune 500 companies in China deploying generative AI by 2025, establishing a leading position in this market segment [5]. Group 3: Technological Advancements - Alibaba's new model architecture, Qwen3-Next, released on September 12, boasts performance ten times that of its predecessor at only one-tenth the construction cost [8]. - Baidu's Wenxin (ERNIE) X1.1, released on September 9, shows significant improvements in authenticity and instruction adherence, comparable to GPT-5 and Gemini 2.5 Pro [9]. - Tencent has launched advanced models such as HunyuanWorld-Voyager and HunyuanImage 2.1, enhancing its capabilities in 3D world generation and high-resolution image generation [10]. Group 4: Market Potential and Challenges - Despite rapid advancements in AI models and applications, commercial monetization in China remains a significant challenge, with only $1.5 billion in annual recurring revenue (ARR) from AI applications as of August 2025, representing just 5% of the global total of $30 billion [12][13].

Core Viewpoint - The rapid adoption of generative AI by Chinese enterprises is driving significant growth in AI infrastructure investments, with Alibaba positioned favorably due to its leading model capabilities and market share [3][4][6]. Group 1: Market Dynamics - The daily token consumption of enterprise-level large models in China is projected to reach 10.2 trillion by the first half of 2025, representing a staggering 363% increase compared to the second half of 2024 [2][6]. - Alibaba, ByteDance, and DeepSeek are the top three choices for Chinese enterprises when selecting general large models [6]. - High demand for AI in commercial applications is accelerating the deployment and usage of AI technologies [3][6]. Group 2: Alibaba's Position - Goldman Sachs has raised Alibaba's target price from $163 to $179, maintaining a "buy" rating, reflecting optimism about the company's growth prospects [4][9]. - Alibaba holds a 47% share of the Chinese public cloud market, which positions it advantageously for international expansion [3][9]. - The company has successfully penetrated most of the Fortune 500 companies in China deploying generative AI by 2025 [6]. Group 3: Infrastructure Investment - The investment narrative for AI infrastructure in China has been reignited following Alibaba's better-than-expected cloud revenue and capital expenditure [8][9]. - Capital expenditures for Chinese cloud service providers are expected to grow by 39% year-on-year by the third quarter of 2025, providing a hardware foundation for sustained AI cloud revenue growth [8][9]. Group 4: Chip Supply and Strategy - The evolution of the chip supply landscape, including a diversified chip supply strategy, is enhancing market confidence and reshaping the development prospects of China's AI cloud industry [9]. - Chinese cloud vendors are making progress in self-developed inference chips, indicating a reduced reliance on overseas chip supplies [4][9]. Group 5: Competitive Landscape - Major players in the AI model and application space are rapidly advancing, with Alibaba releasing its Qwen3-Next model, which boasts ten times the performance of its predecessor at a tenth of the cost [11]. - Other companies like Baidu and Tencent are also making significant strides in AI model development, enhancing their competitive positioning [11][12]. Group 6: Commercialization Challenges - Despite rapid advancements in AI technology, the commercialization of AI applications in China still faces significant challenges, with only $1.5 billion in annual recurring revenue (ARR) compared to $30 billion globally [14].

在近日召开"高德扫街榜"产品发布会上，高德地图对外宣布，其面向出行生活打造的智能体（Agent）"小高老师"单日调用量已达12亿次。 最新数据显示，上线不到一个月，"小高老师"的月活跃用户数已突破4亿，同时它也成为高德连接用户与生活服务的重要载体。目前，高德地图上每 天有1.2亿次生活服务AI搜索，而"小高老师"日均推荐的生活服务类点位超6300万个。 高德地图CEO郭宁表示："我们对地图的理解，正在发生一次根本性的改变。地图不再是一个'你问我答'的工具，而是一个'先于你思考'的伙伴。它将 从用户的手机里'走出来'，成为用户值得信赖的、全场景的出行生活决策助手。同时，它也能将用户的精力从'如何到达'的繁琐中解放出来，真正交 还给生活本身。" 目前，"小高老师"正持续迭代升级。过去一个月，其在行程规划、复杂需求识别和满足等核心能力上取得显著提升。未来，"小高老师"将重点提升自 我反思、异地出游等专项服务能力。 （文章来源：国际金融报） "小高老师"是高德地图于8月初推出的全球首个AI原生地图应用——高德地图2025的核心引擎。在高德地图2025版中，用户通过点击搜索栏里的语音 图标或点击首页下方的"对话"，即可 ...

高德地图CEO郭宁表示："我们对地图的理解，正在发生一次根本性的改变。地图不再是一个'你问我 答'的工具，而是一个'先于你思考'的伙伴。它将从用户的手机里'走出来'，成为用户值得信赖的、全场 景的出行生活决策助手。同时，它也能将用户的精力从'如何到达'的繁琐中解放出来，真正交还给生活 本身。" 新浪科技讯 9月12日晚间消息，在近日召开"高德扫街榜"产品发布会上，高德地图对外宣布，其面向出 行生活打造的智能体（Agent）"小高老师"单日调用量已达12亿次。最新数据显示，上线不到一个 月，"小高老师"的月活跃用户数已突破4亿，同时它也成为高德连接用户与生活服务的重要载体。目 前，高德地图上每天有1.2亿次生活服务AI搜索，而"小高老师"日均推荐的生活服务类点位超6300万 个。 "小高老师"是高德地图于8月初推出的全球首个AI原生地图应用——高德地图2025的核心引擎。在高德 地图2025版中，用户通过点击搜索栏里的语音图标或点击首页下方的"对话"，即可与"小高老师"进行交 流。 目前，"小高老师"正持续迭代升级。过去一个月，其在行程规划、复杂需求识别和满足等核心能力上取 得提升。未来，"小高老师"将重点提升 ...

在近日召开"高德扫街榜"产品发布会上，高德地图对外宣布，其面向出行生活打造的智能体（Agent）"小高老师"单日 调用量已达12亿次。 高德地图CEO郭宁表示："我们对地图的理解，正在发生一次根本性的改变。地图不再是一个'你问我答'的工具，而是 一个'先于你思考'的伙伴。它将从用户的手机里'走出来'，成为用户值得信赖的、全场景的出行生活决策助手。同时， 它也能将用户的精力从'如何到达'的繁琐中解放出来，真正交还给生活本身。" 目前，"小高老师"正持续迭代升级。过去一个月，其在行程规划、复杂需求识别和满足等核心能力上取得显著提升。 未来，"小高老师"将重点提升自我反思、异地出游等专项服务能力。 （文章来源：证券时报网） "小高老师"是高德地图于8月初推出的全球首个AI原生地图应用——高德地图2025的核心引擎。在高德地图2025版中， 用户通过点击搜索栏里的语音图标或点击首页下方的"对话"，即可与"小高老师"进行交流。 无论是出行前，定制多天自驾游攻略、查询机场贵宾厅权益，还是预订私密包间餐厅，亦或是出行中实时调整和更换 出行方案，"小高老师"都能凭借强大的语义理解能力，精准解析需求，在推理思考后调用子智能体和 ...

围绕线下生活，服务消费正迎来蓬勃发展。2024年，我国服务业占GDP的比重持续提升，达到了56.7%；居民人均服务性消费支出占人均消费支出的比重 46.1%。同时，服务业就业人员达3.58亿人，占就业人员总数的48.8%。 服务消费一头连着经济，一头连着民生，但随着服务消费规模不断扩大，一些问题也逐渐显现。今年以来，高德在多场线下调研中发现，无论是商家还是消 费者，都期待一个更加"真实、开放、可信、简单"的消费环境。 郭宁表示，这八个字背后承载着一种更深刻的商业文明，那就是契约精神、合作理性与长期主义。用AI探索打造一套全新线下服务信用体系，正是培育这 种商业文明的最优技术路径，即通过服务信用体系去提升消费信任，提振消费信心，让商家用心经营，让用户安心消费，最终用科技助力服务消费新增长。 9月10日，阿里巴巴旗下高德地图推出全新的AI生活服务榜单"扫街榜"，该榜基于高德地图规模化、真实可信的出行和评价数据，以科学的AI算法模型驱动 生成，覆盖全国超300多个城市的线下吃喝玩乐全品类服务。 "科技既是星辰大海，更是人间烟火。"高德地图CEO郭宁表示，扫街榜的初衷是用科技AI去打造一份真实榜单，通过提升消费信任， ...

本报讯 （记者袁传玺）9月10日，阿里巴巴旗下高德地图推出全新的AI生活服务榜单"扫街榜"，该榜基 于高德地图规模化、真实可信的出行和评价数据，以科学的AI算法模型驱动生成，覆盖全国超过300多 个城市的线下吃喝玩乐全品类服务。 "科技既是星辰大海，更是人间烟火。"高德地图CEO郭宁表示，扫街榜的初衷是用科技AI去打造一份真 实榜单，通过提升消费信任，来提振消费信心，引导更多消费者前往线下放心消费；通过探索AI时代 的线下服务信用体系，助力服务消费新增长。他预计，扫街榜每天将为线下店铺多带去1000万消费者。 据了解，今年8月份，高德地图宣布全面AI化，将地图导航服务演进为具备深度时空理解和自主推理决 策能力的出行生活智能体，成为每一个人的行动好伙伴。 发布会上，郭宁首次披露了高德地图的AI化成果：高德AI助手"小高老师"单日调用量突破12亿次；本地 生活类点位，日均AI搜索量1.2亿次、日均导航总里程15亿公里。这些数字背后，是高德地图运用AI技 术对用户出行和消费习惯的深度理解和重塑。 郭宁表示，用AI探索打造一套全新线下服务信用体系，正是培育这种商业文明的最优技术路径，即通 过服务信用体系去提升消费信任 ...

Core Viewpoint - Alibaba's Gaode Map launched a new AI-driven service ranking called "Street Ranking," aimed at enhancing consumer trust and boosting offline consumption by leveraging real travel and review data across over 300 cities in China [1][2]. Group 1: AI Integration and Consumer Behavior - Gaode Map has fully integrated AI into its services, evolving from a navigation tool to a comprehensive lifestyle service platform, with daily AI search volumes reaching 120 million and navigation distances totaling 1.5 billion kilometers [2]. - The service sector in China is projected to account for 56.7% of GDP in 2024, with per capita service consumption expenditure making up 46.1% of total consumption [2]. Group 2: Trust and Credibility in Consumption - The "Street Ranking" emphasizes "realness," utilizing a large volume of genuine behavioral data to ensure trustworthy evaluations, while also incorporating Sesame Credit scores to filter out false information [4][5]. - The initiative aims to create a more authentic and reliable consumption environment, addressing issues like distorted ratings that affect consumer choices [2][3]. Group 3: Support for Small Businesses - Gaode Map introduced the "Firewood Good Store Support Plan," which includes a budget of 2 billion yuan in ride-hailing vouchers and 9.5 billion yuan in consumption vouchers to drive foot traffic to small businesses [6]. - The plan aims to help 100,000 small stores gain visibility and customer flow, with an expected daily increase of 10 million consumers visiting offline stores due to the "Street Ranking" [5][6].

Core Viewpoint - The article discusses the transformation of Gaode Map into a fully AI-driven service, referred to as "Xiao Gao Teacher," which enhances user experience by providing personalized travel and lifestyle recommendations based on real-time data and user preferences [21][52]. Group 1: Transformation of Gaode Map - Gaode Map has evolved from a simple navigation tool to an intelligent assistant that integrates various aspects of travel and daily life [21][36]. - The introduction of the ST-MAC system allows for multi-agent collaboration, enabling the app to understand and fulfill complex user requests [25][27]. - The AI system can dynamically adjust travel plans based on real-time conditions, such as traffic and user preferences, creating a seamless experience [33][47]. Group 2: User Experience Enhancement - Users can interact with "Xiao Gao Teacher" to plan routes, find dining options, and manage schedules without needing to break down the steps themselves [14][16]. - The system can handle multiple dimensions of user needs, such as location, weather, and real-time traffic, to provide tailored recommendations [28][30]. - The app's ability to learn from user interactions allows it to refine its suggestions over time, enhancing the overall user experience [33][52]. Group 3: Integration of Services - Gaode Map aims to integrate various services, such as transportation, dining, and leisure activities, into a cohesive user experience [36][52]. - The app's architecture allows for the inclusion of third-party services, transforming them into active components of the travel experience [36][52]. - The focus has shifted from merely providing directions to creating a comprehensive service that anticipates user needs and preferences [53][54].