2025年08月25日 证券研究报告 | 人工智能行业专题（11） 探究模型能力与应用的进展和边界 行业研究 · 行业专题 互联网 · 互联网II 投资评级：优于大市（维持） 证券分析师：张伦可 证券分析师：陈淑媛 证券分析师：刘子谭 证券分析师：张昊晨 0755-81982651 021-60375431 liuzitan@guosen.com.cn zhanghaochen1@guosen.com.cn zhanglunke@guosen.com.cn chenshuyuan@guosen.com.cn S0980525060001 S0980525010001 S0980521120004 S0980524030003 请务必阅读正文之后的免责声明及其项下所有内容 报告摘要 Ø 风险提示：宏观经济波动风险、广告增长不及预期风险、行业竞争加剧风险、AI技术进展不及预期风险等。 请务必阅读正文之后的免责声明及其项下所有内容 2 Ø 本篇报告主要针对海内外模型发展、探究模型能力与应用的进展和边界。我们认为当前海外模型呈现差异化发展，企业调用考虑性价比。当前 OpenAI在技术路径上相对领先，聚焦强化推理与专业 ...

Group 1: Key Trends in AI Models - The article highlights various AI models such as Grok 4 by xAI, DeepSeek-R2 by DeepSeek, and GLM-4.1V-Thinking by Zhizhu, showcasing advancements in AI technology [2] - Notable models include Omni-Infer by Huawei, PEVA world model by LeCun team, and Pangu open-source model by Huawei, indicating a competitive landscape in AI model development [2] - Major companies like Google and Tencent are also developing models such as Gemma 3n and Hunyuan-A13B, respectively, reflecting the ongoing innovation in the AI sector [2] Group 2: AI Applications - The article lists various AI applications, including AI game engines by Google and NVIDIA, and Gemini for Education by Google, demonstrating the diverse use cases of AI technology [2][3] - Other applications mentioned are MAI-DxO by Microsoft and AI customization services by OpenAI, indicating a trend towards personalized AI solutions [3] - The introduction of AI-powered tools like GitHub Copilot Chat and document summarization upgrades by Tencent Yuanbao highlights the growing integration of AI in everyday tasks [3] Group 3: Industry Insights and Opinions - The article discusses the impact of AI on employment as noted by the World Economic Forum, suggesting significant changes in job markets due to AI advancements [3] - Perspectives on AI writing influence from The New Yorker and strategic paths from Amazon provide insights into how AI is reshaping industries [3] - The mention of AI economic experiments by Anthropic indicates a focus on understanding the economic implications of AI technologies [3] Group 4: Events and Developments - Key events include the poaching of Claude by Anysphere and new AI crawler regulations by Cloudflare, reflecting the competitive dynamics in the AI industry [4] - The establishment of a superintelligence lab by Meta signifies a push towards advanced AI research and development [4] - The article also notes the talent acquisition efforts by Meta targeting OpenAI, highlighting the ongoing race for top AI talent [4]

Group 1 - The Hong Kong stock market is closed today to celebrate the 28th anniversary of its return, while A-share related ETFs are trading normally, with the Hang Seng Tech ETF (513260) experiencing a slight decline of 0.29% during the day [1] - The Hang Seng Tech ETF (513260) is noted for having the lowest management fee in its category at only 0.15% [1] - As of the last trading day, most constituent stocks of the Hang Seng Tech ETF (513260) have retreated, with Meituan down over 3% and Alibaba down over 2%, while Xiaomi and Kuaishou saw slight increases [3] Group 2 - Southbound funds have continued to net buy Hong Kong stocks, with a total net purchase exceeding HKD 731.1 billion this year, accounting for over 90% of the total net buy for 2024 [4] - As of June 20, the total market value of stocks held by the Hong Kong Stock Connect exceeded HKD 49.394 trillion, with the financial sector holding the highest proportion [4] - The top ten stocks held by southbound funds include Tencent Holdings, China Mobile, and Construction Bank, with Tencent's market value exceeding HKD 500 billion [6][7] Group 3 - Recent net purchases by southbound funds have seen significant inflows into the Hang Seng Tech ETF (513260) constituents, particularly in SMIC, Meituan, and Kuaishou [7] - The estimated net inflow into ETFs in the Hong Kong stock market is approximately HKD 57.557 billion, with significant inflows into the information technology and non-essential consumer sectors [8] - Long-term prospects for the Hong Kong stock market remain strong, with a focus on technology growth and high dividend strategies, despite short-term volatility [11]

6.18 的前一天，在百度科技园一间报告厅里，两段罗永浩直播带货的视频片段正在大屏幕的两侧依 次播放。其中一条棉柔巾的视频里，罗永浩调侃着男性用户也可以跟着媳妇用上棉柔巾了；另一条视 频里，他吐槽了评论里让他打开瘦脸功能的建议，说自己不是颜值主播。 这两条视频哪个是数字人？百度副总裁、百度电商总经理平晓黎让在场的媒体猜猜看，现场的投票几 乎是平手。 不过，最终的结果出乎大家预料 —— 两段视频其实都来自罗永浩的数字人。 从标志性的 " 罗氏幽默 " ，到语气停顿节奏和微表情的精准捕捉，数字人对罗永浩的复刻几乎达到 以假乱真的突破。以至于当百度随后播放了一段关于罗永浩谈数字人直播感受的真人视频时，不少人 仍下意识怀疑这是 AI 生成的内容。 6 月 15 日，罗永浩数字人在百度电商完成首场直播，吸引超 1300 万人次观看。 GMV 突破 5500 万元，部分 3C 、食品等核心品类商品带货单量超罗永浩 5 月真人首秀同期数据，创下数字 人直播带货新纪录。 在电商这个讲究 " 即时转化 " 的残酷战场，能卖货的 AI 才是真实力。这场直播不仅验证了百度数 字人技术的商业价值，更揭示了 AI 大模型正从技术秀场转 ...

Core Viewpoint - MiniMax has launched the M1 model, claiming it to be the world's first open-source large-scale hybrid architecture inference model, which significantly improves long text processing capabilities and reduces reinforcement learning costs [1][3][6] Pricing Strategy - The pricing for the M1 model is structured as follows: - For 0 to 32,000 tokens: 0.8 RMB per million tokens for input and 8 RMB for output - For 32,000 to 128,000 tokens: 1.2 RMB per million tokens for input and 16 RMB for output - For 128,000 to 1,000,000 tokens: 2.4 RMB per million tokens for input and 24 RMB for output - The first two pricing tiers are lower than those of DeepSeek-R1, while the third tier covers a market segment not yet addressed by competitors [4][3] Model Capabilities - The M1 model supports a context window of up to 1,000,000 tokens, matching Google's Gemini 2.5 Pro and being nearly eight times the capacity of DeepSeek-R1, which supports 128,000 tokens [4][3] - The model also features the longest inference output capability in the industry at 80,000 tokens [4] Technological Innovation - The breakthrough in M1's performance is attributed to the unique "Lightning Attention" hybrid architecture, which optimizes the attention mechanism for long sequences, addressing the computational bottleneck of traditional transformer models [6] - The introduction of the CISPO algorithm enhances the efficiency and stability of reinforcement learning by optimizing the importance sampling weights rather than adjusting token updates [6] Market Positioning - MiniMax's focus is on accelerating technological iteration rather than immediate revenue growth, indicating a strategic shift in response to competition from DeepSeek and other players in the AI space [8][10] - The company aims to leverage its technological advancements to position itself favorably in the emerging AI Agent market, which is expected to evolve significantly by 2025 [9][10] Competitive Landscape - The launch of M1 comes amid a competitive environment where other companies like Alibaba and Baidu are also releasing advanced models with competitive pricing and capabilities [7][9] - Industry experts suggest that while MiniMax's advancements are notable, the true commercial value will depend on user feedback and practical applications [7][10]

Core Insights - Baidu's Vice President Wu Tian presented the latest technological innovations of the Wenxin large model during the Baidu AI Day event on May 20 [1] - The Wenxin model 4.5 is a multimodal foundational large model, with Wenxin 4.5 Turbo derived from it, offering better performance at a lower cost [1] - Based on Wenxin 4.5 Turbo, the Wenxin X1 has been upgraded to X1 Turbo, enhancing performance while incorporating more advanced reasoning chains, with improved capabilities in Q&A, content creation, logical reasoning, tool invocation, and multimodal functions [1]

Group 1 - The core viewpoint of the articles is the advancement of Baidu's AI technologies, particularly the release of the Wenxin large model 4.5 and the X1 Turbo model, which enhance multi-modal capabilities and problem-solving efficiency [1][2] - Wenxin 4.5 Turbo improves performance and reduces costs compared to Wenxin 4.5, while the X1 Turbo model incorporates advanced thinking chains for better question answering, creation, logical reasoning, and tool utilization [1] - Multi-modal capabilities of Wenxin 4.5 and 4.5 Turbo include mixed training of text, images, and videos, achieving nearly double the learning efficiency and over 30% improvement in multi-modal understanding [1] Group 2 - The deep thinking aspect of Wenxin X1 and X1 Turbo combines tool utilization with thinking paths, creating a composite thinking chain that enhances output quality [1] - Baidu's ultra-realistic digital human technology utilizes multi-modal AI to ensure coherence in language, voice, and appearance, currently supporting over 100,000 digital human hosts with a 31% live conversion rate and an 80% reduction in live broadcast costs [4] - The Wenxin Fast Code assistant generates over 40% of new code daily, serving 7.6 million developers, while the training throughput of Wenxin 4.5 Turbo is 5.4 times that of Wenxin 4.5 [4]

Core Insights - The article highlights significant developments in the AI industry, including major investments, product launches, and strategic shifts by leading companies [2][24]. Domestic Developments - Alibaba is reportedly aiming for full "AI integration" across its business by 2025, with all departments evaluated on their AI utilization for growth [4][5]. - Notable investor Duan Yongping purchased $9.25 million worth of Nvidia shares, but Nvidia's stock fell by 3.43% on the day of the announcement [5]. - Manus has registered its generative AI service Monica in Beijing, with the city leading in the number of registered large model products [5]. - The Step-Video-TI2V model was released, capable of generating 5-second videos with controllable motion [5]. - Yushutech showcased its humanoid robot and quadruped robot at AWE2025, attracting significant attention [6]. - The AI service Deepseek will assist users in estimating repair costs, enhancing transparency in pricing [8]. - Xiaomi's AI search and writing tools were among 34 newly registered generative AI services in Beijing [14]. International Developments - OpenAI launched new voice models aimed at developing voice AI agents, marking a significant advancement in the field [14]. - Elon Musk's xAI is collaborating with BlackRock to establish an AI infrastructure investment fund, indicating a competitive stance against OpenAI [15]. - Apple is restructuring its AI leadership to revitalize its AI initiatives, particularly focusing on improving Siri [15][16]. - Nvidia introduced the GROOT N1, an open-source humanoid robot model, which is expected to accelerate the development of humanoid robots [19][21]. - Google announced a $32 billion acquisition of cloud security company Wiz, enhancing its cloud capabilities in the AI era [21]. Investment and Financing Overview - This week, there were 7 disclosed AI financing events globally, totaling approximately 563 million RMB, with an average investment of 94 million RMB [25][31]. - The majority of domestic AI financing events were concentrated in Zhejiang and Beijing, with 4 and 3 events respectively [28]. - Zhihui AI, a developer of AI knowledge technology, completed a 300 million RMB D++ round financing, focusing on large model innovations [31].

资料图。 本文来源：财联社、上观新闻 百度安全负责人："谢广军女儿开盒事件"数据并 非从百度泄露 网暴素人孕妇？ 网友怀疑其利用"百度数据库"获取网友隐私 虽然谢广军已道歉，但该事件还有诸多传言疑云未解。 据报道，微博网友@你的眼眸是世界上最小的湖泊（以下简称"眼眸"）自2024年以来，多次 在饭圈"开盒"素人。 有消息称，3月上旬，其参与对一名孕妇网友的"开盒"网暴，同多人一起曝光对方工作单位 并私信其丈夫进行辱骂，引发众多关注。 财联社3月18日电，百度安全负责人陈洋在公司内网做出了一段回应说明，表示在接到举报 后，已经展开了调查并完成取证，结果显示，"谢广军女儿开盒事件"数据并非从百度泄露。 其中提到两点：一是百度内部做了数据的匿名化、假名化处理，并做了权限分离，没有人能 有权限拿到数据，谢广军也没有数据库的任何相关权限。二是通过对此次开盒事件脉络的模 拟调查，发现可以在海外社交媒体群里找到大量个人信息，而且很多信息都是免费的。 百度副总裁为女儿道歉后，网友质疑：是否利用百 度数据库获取隐私 据第一财经报道 ， 针对近日网络传闻， 百度副总裁谢广军 发布微信朋友圈 表示深感愧疚， 并向所有受影响的朋友郑 ...

热门板块方面 盘面上，大型科技股全线飘红，百度劲升12%，阿里巴巴、京东均涨超5%；机器人概念股午后拉 升，越疆大涨超27%，其具身智能人形机器人正式预售；CRO概念股走势强劲，药明康德绩后涨超 10%；国际金价再创历史新高，黄金股集体上涨；汽车股普遍走强，蔚来最高涨超17%，宁德时代 （300750）正推进对蔚来能源不超过25亿元的战略投资；AI概念股、芯片股、苹果概念股等悉数走 高。另一边，此前表现亮眼的乳业股多数走低。 1. AI概念股走势强劲。截至收盘，百度集团-SW(09888)涨12.22%，报103.3港元；金山云(03896)涨 10.86%，报9.7港元；阿里巴巴-W(09988)涨5.83%，报143.4港元；万国数据-SW(09698)涨5.56%，报35.1 港元。 港股收盘 | 恒指收涨2.46% 机器人、AI概念股走强 百度集团-SW劲升逾12% 港股三大指数全天走强，恒指收涨595点刷新三年新高，恒科指数则大涨近4%。截止收盘，恒生指 数涨2.46%或595点，报24740.57点，全日成交额2859.35亿港元；恒生国企指数涨2.79%，报9177.8点； 恒生科技指数涨3.9 ...