证券研究报告 2025年07月14日 海外行业报告：行业动态研究 英伟达股价创新高，xAI发布Grok 4系列模型 全球AI周报 作者： 分析师 孔蓉 SAC执业证书编号：S1110521020002 分析师 李泽宇 SAC执业证书编号：S1110520110002 分析师 樊程安吉 SAC执业证书编号：S1110524080001 分析师 杨雨辰 SAC执业证书编号：S1110521110001 分析师 刘诗雨 SAC执业证书编号：S1110524120001 请务必阅读正文之后的信息披露和免责申明 ◼ 全球AI动态： ◼ 投资建议： 请务必阅读正文之后的信息披露和免责申明 2 ➢ xAI发布Grok 4系列模型：推理能力升级，定价超OpenAI。马斯克旗下xAI发布Grok 4系列模型，包括单代理版本的Grok 4和支持四个代理同时工作的多代理版本Grok 4 Heavy，两者均为纯推理模型，上下文窗口最 高支持256k tokens。Grok 4 Heavy在HLE测试中得分44.4%，超过谷歌Gemini 2.5 Pro，在GPQA、AIME25等测试中也表现优异，且训练量是Grok 2的100倍，G ...

2.【Hugging Face开源小参数模型SmolLM3】北京时间7月9日凌晨，Hugging Face首席执行官克莱门特·德朗 格（Clement Delangue）宣布，Hugging Face发布并开源小参数模型SmolLM3。拥有128k上下文窗口；支持 英语、法语、西班牙语、德语等6种语言；支持深度思考和非思考双推理模式。（财联社） 3.【浙江大学发布高精准基因组设计AI模型】浙江大学郭国骥教授团队开发出一款用于基因组预测设计的深度学 习AI模型"女娲CE"，能够以超过90%的准确率预测基因组调控区域发生突变之后带来的表型变化，并结合疾病 表型设计出相应的治疗位点。相关成果已发表于国际学术期刊《细胞》。 （财联社） 4.【Hugging Face 桌面机器人 Reachy Mini 开订：长相呆萌，支持超 170 万个 AI 模型】据外媒TechCrunch 报道，Hugging Face旗下最新桌面机器人Reachy Mini的订单现已正式开放，开发者现在已可动手组装与测试。 Reachy Mini将推出两个版本。无线版名为Reachy Mini Wireless，内置Raspberry 5 微 ...

Group 1: Veo 3 Upgrade - The Google Veo 3 upgrade allows audio and video generation from a single image, maintaining high consistency across multiple angles [1] - The new feature is implemented through the Flow platform's "Frames to Video" option, enhancing camera movement capabilities, although the Gemini Veo3 entry is currently unavailable [1] - User tests indicate natural expressions and effective performances, marking a significant breakthrough in AI storytelling applicable in advertising and animation [1] Group 2: Hugging Face 3B Model - Hugging Face has released the open-source 3B parameter model SmolLM3, outperforming Llama-3.2-3B and Qwen2.5-3B, supporting a 128K context window and six languages [2] - The model features a dual-mode system allowing users to switch between deep thinking and non-thinking modes [2] - It employs a three-stage mixed training strategy, trained on 11.2 trillion tokens, with all technical details, including architecture and data mixing methods, made available [2] Group 3: Kunlun Wanwei Skywork-R1V 3.0 - Kunlun Wanwei has open-sourced the Skywork-R1V 3.0 multimodal model, achieving a score of 142 in high school mathematics and 76 in MMMU evaluation, surpassing some closed-source models [3] - The model utilizes a reinforcement learning strategy (GRPO) and key entropy-driven mechanisms, achieving high performance with only 12,000 supervised samples and 13,000 reinforcement learning samples [3] - It excels in physical reasoning, logical reasoning, and mathematical problem-solving, setting a new performance benchmark for open-source models and demonstrating cross-disciplinary generalization capabilities [3] Group 4: Vidu Q1 Video Creation - Vidu Q1's multi-reference video feature allows users to upload up to seven reference images, enabling strong character consistency and zero storyboard video generation [4] - Users can combine multiple subjects with simple prompts, with clarity upgraded to 1080P, and support for character material storage for repeated use [5] - Test results show it is suitable for creating multi-character animation trailers, supporting frame extraction and quality enhancement, reducing video production costs to less than 0.9 yuan per video [5] Group 5: VIVO BlueLM-2.5-3B Model - VIVO has launched the BlueLM-2.5-3B edge multimodal model, which excels in over 20 evaluations and supports GUI interface understanding [6] - The model allows flexible switching between long and short thinking modes, introducing a thinking budget control mechanism to optimize reasoning depth and computational cost [6] - It employs a sophisticated structure (ViT+Adapter+LLM) and a four-stage pre-training strategy, enhancing efficiency and mitigating the text capability forgetting issue in multimodal models [6] Group 6: DeepSeek-R1 System - The X-Masters system, developed by Shanghai Jiao Tong University and DeepMind Technology, has achieved a score of 32.1 in the "Human Last Exam" (HLE), surpassing OpenAI and Google [7] - The system is built on the DeepSeek-R1 model, enabling smooth transitions between internal reasoning and external tool usage, using code as an interactive language [7] - X-Masters employs a decentralized-stacked multi-agent workflow, enhancing reasoning breadth and depth through collaboration among solvers, critics, rewriters, and selectors, with the solution fully open-sourced [7] Group 7: Zhihui Jun's Acquisition - Zhihui Jun's Zhiyuan Robot has acquired control of the listed company Shuangwei New Materials for 2.1 billion yuan, aiming for a 63.62%-66.99% stake [8] - Following the acquisition, Shuangwei New Materials' stock resumed trading with a limit-up, reaching a market value of 3.77 billion yuan, with the actual controller changing to Zhiyuan CEO Deng Taihua and core team members including "Zhihui Jun" Peng Zhihui [8] - This acquisition, conducted through "agreement transfer + active invitation," is seen as a landmark case for new productivity enterprises in A-shares following the implementation of national policies [8] Group 8: AI Model Usage Trends - In the first half of 2025, the Gemini series models captured nearly half of the large model API market, with Google leading at 43.1%, followed by DeepSeek and Anthropic at 19.6% and 18.4% respectively [9] - DeepSeek V3 has maintained a high user retention rate since its launch, ranking among the top five in usage, while OpenAI's model usage has fluctuated significantly [9] - The competitive landscape shows differentiation: Claude-Sonnet-4 leads in programming (44.5%), Gemini-2.0-Flash excels in translation, GPT-4o leads in marketing (32.5%), and role-playing remains highly fragmented [9] Group 9: AI User Trends - A report by Menlo Ventures indicates that there are 1.8 billion AI users globally, with a low paid user rate of only 3%, and a high student usage rate of 85%, while parents are becoming heavy users [10] - AI is primarily used for email writing (19%), researching topics of interest (18%), and managing to-do lists (18%), with no single task dependency exceeding one-fifth [10] - The next 18-24 months are expected to see six major trends in AI: rise of vertical tools, complete process automation, multi-person collaboration, explosion of voice AI, physical AI in households, and diversification of business models [10]

据昆仑万维（300418）官微消息，7月9日，昆仑万维发布并开源Skywork-R1V3.0版本，其在后训练阶 段通过强化学习策略深度激发模型的跨模态推理能力，在复杂逻辑建模与跨学科泛化方面实现双重飞 跃。据悉，昆仑万维目前已全面开源Skywork-R1V3.0的所有资源。 ...

生成式AI 一、 Grok 4逆天跑分泄露，「人类最后考试」豪取45%全场第一 ？ 1. Grok 4在「人类最后考试」(HLE)测试中得分高达45%，远超Gemini 2.5 Pro和Claude 4 Opus， 引发 讨论 ； 2. 马斯克表示Grok 4以「第一性原理」构建推理机制，像物理学家那样思考，从基本公理层 面分析问题； 3. Grok 4将强化编码能力， 或 分为Grok 4和Grok 4 Code两个版本，预计在7月4日后随时 发布。 https://mp.weixin.qq.com/s/kuk8MfUW_wbS5RAOdV24ZA 二、 Gemini CLI 重磅更新：将 支持音视频处理，与 多项体验升级 1. Gemini CLI 发 布 更新支持音视频输入功能，显著扩展多模态交互能力 ； 实则 目前仅能 处理文本、图片和PDF文件； 2. 增强Markdown功能，新增表格渲染与文件导入功能，并集成VSCodium和Neovim编辑 器，提升开发体验； 3. 技术栈升级至Ink 6和React 19，添加新主题、隐私管理功能，并优化历史记录压缩算法， 提高性能和稳定性。 四、 开源De ...

机器之心报道 编辑：杜伟、泽南 大语言模型（LLM）以生成能力强而著称，但如何能让它「听话」，是一门很深的学问。 基于人类反馈的强化学习（RLHF）就是用来解决这个问题的，其中的奖励模型 （Reward Model, RM）扮演着重要的裁判作用，它专门负责给 LLM 生成 的内容打分，告诉模型什么是好，什么是不好，可以保证大模型的「三观」正确。 因此，奖励模型对大模型能力来说举足轻重：它既需要能够准确进行评判，又需要足够通用化，覆盖多个知识领域，还需要具备灵活的判断能力，可以处理 多种输入，并具备足够的可扩展性。 7 月 4 日，国内 AI 科技公司昆仑万维发布了新一代奖励模型 Skywork-Reward-V2 系列，把这项技术的上限再次提升了一截。 Skywork-Reward-V2 系列共包含 8 个基于不同基座模型和不同大小的奖励模型，参数规模从 6 亿到 80 亿不等，它在七大主流奖励模型评测榜单上全部 获得了第一。 Skywork-Reward-V2 系列模型在主流基准上的成绩。 与此同时，该系列模型展现出了广泛的适用性，它在多个能力维度上表现出色，包括对人类偏好的通用对齐、客观正确性、安全性、风 ...

本周AI商业焦点必读 本周核心趋势 2025年6月24日 1、 「功能扩展」 腾讯元宝、DeepSeek、豆包掀起AI编程新玩法 本质变革：AI将"需求→代码"链路压缩至一句话指令，传统编程工具和低代码平台遭降维打击！ 全栈覆盖 （2025.6.17-6.24） 具身智能产业化加速： 制造巨头纷纷押注具身智能机器人领域，以不断提升"机器服务密度"的 方式逐步替代传统人力配置，从而重新配置其生产业务线的人力资源结构。 多模态进入成本血拼阶段： 随着视频生成成本骤降和开源模型大量涌现，开源方案正强势重构 整个创作生态。 AI算法陪伴重构消费逻辑： "孤独经济"持续催生陪伴型消费需求，Z世代表现出强烈的付费意 愿，愿意为具有仿生记忆功能的AI陪伴服务买单。 自然语言终结传统开发模式： AI编程工具正深刻重构产业链，将复杂的"需求→代码"流程高度 压缩为直观的对话指令；低代码平台的固有价值因此被显著削弱，开发范式正加速向意图层迁 移。 交互革命 场景屠杀 原文链接： 啊？豆包居然也开始卷AI编程了？ 元宝搭载DeepSeek V3实现10+语言编程（Python/Java/C++），0配置实时运行 DeepSite ...

机器之心报道 编辑：Panda 400 多年前，宋应星著成《天工开物》。这是一部写给匠人、也写给未来的书。它让人相信：技术不是死物，而是人与世界持续互动的方式。 如今，代码系统早已成为现代文明的骨架。它们运行在日常软件、银行服务、交通调度等各式系统中，也支撑着我们所依赖的 AI 算法本身。但和古代器物一样， 再精妙的程序也难免出现 bug—— 有些是逻辑失误，有些是环境变迁，有些甚至源于协作失控。比如，就在前几天，AWS、谷歌云、Azure 和 Cloudflare 都发生了 中断，连带着 ChatGPT 和 Cursor 等热门 AI 应用也一并短暂失联；而这一事故的原因可能是一次错误的自动配额更新导致谷歌的 API 管理系统出现了故障。 同时，bug 修复也是软件工程中最基础，却也是最复杂、最消耗人力的任务之一。特别是在真实的 GitHub 项目中，修一个 bug 并不是「找到一行错字那么简 单」，它常常需要： 那么，我们能否使用 AI 智能体来完成这些任务呢？ 当然可以！但我们需要的绝不是传统的用于解决单独编程任务的 AI 编程模型，而是需要像人类开发者一样能够理解历史上下文、进行多轮推理、在模糊与不确 ...

Core Insights - The comprehensive strength rating of Chinese AIGC listed companies can be assessed from various dimensions including technical capabilities, commercialization progress, financial performance, and industry influence [1]. Group 1: Leading Comprehensive AI Companies - Kunlun Wanwei (SZ) has developed the "Tiangong" series of large models covering multiple modalities, achieving global open-source SOTA level with the Skywork-RV model [3]. - Kunlun Wanwei's AI short drama platform DramaWave is projected to generate over 100 million USD in annual revenue by 2025, with AI music business ARR reaching 10 million USD, and a year-on-year revenue growth of 60% to 1 billion CNY in 2025 [4]. - The company is rated as a leader in both technical strength and commercialization capabilities [4]. - Wanjun Technology (SZ) launched the first large model in the digital creative field, "Wanjun Tianmu," supporting real-time video generation in Chinese [5]. - The company is expected to achieve revenue and net profit growth in 2025, with a global presence across more than 20 countries and regions [6]. - Wanjun Technology is recognized as a benchmark for application layer innovation and a leader in the video generation sector [7]. - iFLYTEK (SZ) has a strong foundation in voice recognition and AIGC technology, with a net profit growth rate of 60% year-on-year as per the third-quarter report [8]. - The company has extensive applications in vertical fields such as education and healthcare but needs to enhance multi-modal technology integration [8]. - iFLYTEK is rated as a leader in vertical fields, although its commercialization speed is slightly behind comprehensive players [8]. Group 2: Leading Companies in Vertical Fields - Wisdom Interconnect (AICT) has developed a multi-modal large model (IRN-MMGPT) for intelligent road network applications, serving over 100 million vehicle trips by 2025 [9]. - The company is rated as the absolute leader in the smart transportation sector, with a close integration of technology and application scenarios [9]. - Haitai Ruisheng (SH) has the highest dividend payout capability with a dividend rate of 60%, but its business scale is relatively small [10]. - The company is recognized for its significant advantages in the data service niche, although its overall influence is limited [10]. Group 3: Traditional Companies Transforming - China Telecom (SH) ranks second in revenue (10 billion CNY) and net profit (1 billion CNY) among AIGC concept companies [11]. - The company is making progress in computing infrastructure and industry large models, but AIGC business accounts for a low proportion of its overall operations [11]. - China Telecom is rated for its outstanding resource endowment, though it needs to strengthen its technical barriers [11]. - Midea Group (SZ) ranks third in revenue (8 billion CNY) and second in ROE (60%) among AIGC concept companies [12]. - The company is focusing on integrating industrial robots with smart home scenarios, with AIGC applications still in the early stages [12]. - Midea Group is recognized as a benchmark for digital transformation in manufacturing, with potential yet to be fully realized [12]. Group 4: Emerging Potential Companies - Rongzhi Technology has developed a vertical large model for healthcare, "Dr.GPT," which has won the industry's only award, integrating multi-modal interaction with clinical reasoning capabilities [13]. - The company is rated as a breakthrough player in the healthcare sector, although it needs to expand its commercialization scale [13]. - Jiaodong Technology (SZ) boasts the highest net profit margin (60%) and cost-to-profit ratio (60%) in the industry [14]. - The company is recognized as a model of high profitability, although it has a high business concentration [14]. Group 5: Rating Summary - Kunlun Wanwei is rated as the top tier for its full-stack technology and commercialization breakthroughs [15]. - Wanjun Technology is recognized as a leader in the video generation field and a benchmark for application layers [15]. - Wisdom Interconnect is rated as the top player in the vertical field of smart transportation with its multi-modal large model [15]. - China Telecom and Midea Group are noted for their scale and resource endowment, showcasing potential in traditional transformation [15].

Core Insights - Quark has launched China's first AI model specifically designed for college entrance examination (Gaokao) application scenarios, featuring three core functions: deep search, application report, and intelligent application selection [3] - The model was fine-tuned by benchmarking against industry experts, ensuring that it meets the unique requirements of Gaokao application, which demands high accuracy and logical coherence [3][4] - The fine-tuning process involved creating a structured knowledge base covering over 2,900 universities and nearly 1,600 undergraduate programs, with a focus on data verification and cross-referencing [4] Model Fine-Tuning - The fine-tuning of the model is critical, involving targeted instruction adjustments and the collaboration of hundreds of experts to create a unique generation mechanism [4] - The team distilled thousands of past decisions from human experts to develop a reasoning chain that informs the model's decision-making process [4] - The model's design includes a mechanism to prevent "hallucinations," ensuring that the final recommendations are based on real data and historical validation [4] Agent Product Development - Quark's team served over 30 million users, with 50% being students from third-tier cities or below, indicating a significant market reach [6] - The introduction of the Agent format represents a shift in the industry, with a focus on providing expert-level advice rather than just public information [6] - The year is seen as a pivotal moment for Agent products, with major tech companies rapidly advancing their offerings in this space [6][7] Market Trends - The global AIGC technology penetration is projected to exceed 40% by 2025, with the AI Agent market expected to grow from $5.1 billion in 2024 to $47.1 billion by 2030, reflecting a compound annual growth rate of 44.8% [7]