出品｜虎嗅黄青春频道 作者｜商业消费主笔 黄青春 头图｜视觉中国 字节跳动 CEO 梁汝波首次公开站台，给了豆包。 6 月 11 日，字节跳动旗下火山引擎开了一场发布会，现场数位拿着号码牌的媒体硬是因为主会场人数爆满被拒之门外 20分钟，即便字节跳动公关竭力与现 场安保交涉两轮，对讲机那头的负责人仍然不为所动，严格遵守出一进一的规则，导致主会场内很多火山员工为了给媒体腾位置都被迫中途出会场协同办 公。 即便第三次交涉后虎嗅有幸进入内场，一番闪转腾挪仍被摩肩接踵的人群堵在了会场最后排的摄像臂旁，仿佛挤进了一节北京早高峰地铁车厢，上一次如此 夸张的阵仗还是年初春运赶高铁（不由感慨，时代抛弃你的时候，连发布会都挤不进去）。 活动开始，无法进入主会场的人群 为什么一场产品发布会搞得这么火爆？ 一方面，DeepSeek 凭一己之力掀翻了互联网，从微信到百度，从美团到小红书，国民级应用纷纷接入 DeepSeek，唯独豆包至今依然坚持自研，且字节系大 模型雨后春笋般冒出来，还能始终保持超高的市场声量，自然会牵动着从业者乃至媒体、客户的神经。 比如，发布会上亮相的豆包1.6-thinking 在复杂推理、竞赛级数学、多轮对话 ...

火山引擎总裁 谭待 "规模决定一切，最终牌桌上一定有火山。" 文 丨 贺乾明 在字节负责云计算业务火山引擎 5 年，谭待变得更从容了。要来这里时，很多朋友都劝他慎重：字节 比同行晚了 10 年才做云计算，肯定起不来。那是他压力最大的阶段。他说， 现在朋友们谈起火山引 擎 ，会说做得还可以。 火山引擎交出了亮眼的成绩单。据市场调研机构 IDC 统计，去年国内云端大模型调用量，火山引擎一 家占了 46.4%，比二三名加起来都多。去年 12 月到现在，豆包大模型平均每天处理的 Token 处理量 翻了两番到 16.4 万亿，已经逼近国外的科技巨头。 谭待说，除了火山引擎用技术优化压低了大模型价格，豆包 App 也帮了不少忙，"很多客户都会拿着 跟豆包的聊天 Case 找过来，问火山引擎能不能在企业场景中实现类似的效果？" 他判断 Agent 是火山引擎更大的机会：过去的云平台为网页和 App 开发设计，需要程序员理解业务， 定义规则和算法，然后驱动数据做事，上限就是所有人的时间总和。现在 Agent 成为主体，模型可以 自己思考、规划、编排流程，自动执行解决问题，上限远高于 App 时代，"我们正在加码加班做好准 ...

火山引擎总裁 谭待 "规模决定一切，最终牌桌上一定有火山。" 文 丨 贺乾明 在字节负责云计算业务火山引擎 5 年，谭待变得更从容了。要来这里时，很多朋友都劝他慎重：字节 比同行晚了 10 年才做云计算，肯定起不来。那是他压力最大的阶段。他说， 现在朋友们谈起火山引 擎 ，会说做得还可以。 6 月 11 日，火山引擎举办春季 Force 原动力大会，推出 MCP 服务、提示词工具（Prompt Pilot）、 知识管理系统、强化学习框架（veRL）等为 Agent 定制开发的套件，并宣布根据 "输入长度" 区间给 Agent 调用的思考模型定价，"综合成本只有豆包 1.5 深度思考模型或 DeepSeek-R1 的三分之一"。 "模型使用成本要降下来，才能推动 Agent 的规模化应用。" 谭待说，经营云平台最重要的是规模要 大，"规模大了，服务器多、负载高，就会对技术有很高的要求，会强迫我们做好各种事情。" 这个判断也让他敢在 2021 就定下 1000 亿元的年收入目标，那年中国云平台第一的阿里云年收入只有 600 亿，大模型浪潮还没有踪影。给他信心的是，算上字节整个集团的业务，火山引擎能调度的计算 资源 ...

Core Viewpoint - The partnership between Guocheng Investment Advisory and Alibaba Cloud marks a significant step towards the digital transformation of the financial industry, leveraging Serverless technology to enhance investment advisory services and drive industry innovation [1][19]. Group 1: Technological Revolution - Guocheng Investment Advisory, established in 2004, is committed to providing comprehensive internet products for small and medium investors, and has launched its intelligent investment advisory software "Decision Maker" in 2024, indicating a key milestone in its digital transformation [2]. - The collaboration with Alibaba Cloud initiates a revolutionary upgrade of the technical architecture, transitioning from traditional operations to a modern cloud-native architecture using Alibaba Cloud's Serverless Application Engine (SAE) [2][3]. Group 2: Serverless Technology Empowerment - The adoption of Alibaba Cloud's Serverless Application Engine (SAE) allows Guocheng Investment Advisory to flexibly respond to business peaks without the need for capacity forecasting or manual scaling, enhancing system responsiveness and resource utilization [3][17]. - The migration of existing business systems to the SAE platform enables automatic scaling and resource elasticity, significantly improving operational efficiency and reducing costs associated with idle resources [4][5][17]. Group 3: API Gateway and Microservices Governance - The introduction of Alibaba Cloud's cloud-native API gateway replaces the previous self-built gateway, providing a unified entry point for external services and enhancing API management capabilities [7][19]. - The integration of Alibaba Cloud's Microservices Engine (MSE) optimizes the microservices architecture, improving service registration, discovery, and governance capabilities, which enhances system robustness [10][12]. Group 4: Observability and DevOps Automation - Guocheng Investment Advisory establishes a comprehensive observability system using Alibaba Cloud's ARMS and SLS services, allowing real-time monitoring of application performance and faster fault detection [13][18]. - The implementation of an automated DevOps system streamlines the release process, reducing errors and improving deployment efficiency through a complete CI/CD pipeline [14][19]. Group 5: Business Value and Future Prospects - The transition to a cloud-native Serverless architecture results in significant cost savings and operational efficiency, with overall resource utilization improving by nearly 50% [17]. - The partnership with Alibaba Cloud not only enhances system stability and reduces operational costs but also lays a solid foundation for future digital innovation and intelligent upgrades, with plans to integrate AI technologies by 2025 [19].

[Table_Main] 证券研究报告 | 行业点评 计算机 2025 年 06 月 12 日 计算机 优于大市（维持） 证券分析师 相关研究 1.《稳定币法案&Circle IPO，关注全 球货币变局机遇》，2025.6.9 2.《海光拟吸收合并曙光，关注中科 系上市公司隐含价值和协同价值》， 2025.5.26 陈涵泊 资格编号：S0120524040004 邮箱：chenhb3＠tebon.com.cn 李佩京 资格编号：S0120524090004 邮箱：lipj＠tebon.com.cn 研究助理 王思 邮箱：wangsi＠tebon.com.cn 市场表现 -27% -13% 0% 13% 27% 40% 54% 67% 81% 2024-06 2024-10 2025-02 计算机 沪深300 资料来源：德邦研究所，聚源数据 3.《博睿数据：拥抱华为与字节，新 产品或进放量周期》，2025.5.12 4.《中科星图：对标 PLTR，AI+数据 分析之王》，2025.5.8 5.《24 年报&25Q1 季报总结： 盈利 反转之年，AI 产业率先体现》， 2025.5.7 火山引擎 Force 原动 ...

机器之心报道 编辑：Panda 高考余热尚在，依然还是有不少博主和媒体在测试各家 AI 模型解答最新高考题的能力。而现在，一个正被火热评测的主流模型迎来了重磅升级！ 在 AI 基础设施方面，火山引擎围绕 AI 应用的实际需求，打造了三个套件： AgentKit 、 TrainingKit 和 ServingKit 。顾名思义，这三个套件的作用分别是智能 体、模型训练和模型 serving。 整场 Force 大会下来，我们最大的感受是火山引擎对「 AI 云原生 」概念的反复强调。简单来说，「AI 云原生」是指将云原生理念应用到 AI 领域，专注于 AI 工 作负载的云端构建和部署；其核心理念是以 AI 负载为中心，重新优化计算、存储与网络架构，让 GPU 可直接访问存储和数据库，降低 IO 延迟，并提供高速互联 和端到端安全保护。事实上，正如我们以前多次报道的那样，「AI 云原生」已经成为火山引擎「下一个十年」所遵循的云计算新范式。 事情发生在昨天、北京国家会议中心、火山引擎原动力大会 Forc e 202 5 现场。 火山引擎一口气发布了豆包大模型 1.6、Seedance 1.0 Pro、AI 云原生平台 ...

Core Insights - The article discusses the launch of various AI models and services by Volcano Engine at the FORCE Original Power Conference, highlighting advancements in the Doubao model series and the introduction of new video and voice generation models [2][31]. Group 1: Doubao Model Series - The Doubao model 1.6 series includes three models: doubao-seed-1.6, 1.6-thinking, and 1.6-flash, with 1.6-thinking surpassing DeepSeek-R1-0528 in reasoning capabilities [2][8]. - Pricing for Doubao 1.6 is based on input length, with costs significantly reduced to 0.8 yuan per million tokens for input and 8 yuan per million tokens for output, making it one-third the cost of previous models [2][5]. - Doubao model 1.6 supports multi-modal understanding and GUI operations, enabling applications in e-commerce, autonomous driving, and security inspections [10][12]. Group 2: Seedance Video Generation Model - The Seedance 1.0 pro model achieved top rankings in text-to-video and image-to-video tasks, outperforming competitors like Veo3 and Kuaishou 2.0 [5][17]. - The cost of generating a 5-second 1080P video is only 3.67 yuan, allowing for the production of over 2,700 videos within a budget of 10,000 yuan, which is claimed to be the lowest in the industry [5][16]. Group 3: AI Cloud Native Services - Volcano Engine upgraded its AI cloud-native services, introducing tools like MCP service, PromptPilot, and AI knowledge management systems, aimed at enhancing agent development and application [7][21]. - The daily token usage for the Doubao model exceeds 16.4 trillion, reflecting a 137-fold increase since its initial release [7][14]. - The Doubao model holds a 46.4% market share in China's public cloud large model market, ranking first [7]. Group 4: Agent Development Tools - Twelve new agent development tools were introduced, including TRAE for code assistance and MCP service for cloud service integration, which aims to streamline the development process [21][22]. - The PromptPilot application optimizes user prompts and supports complex task management, enhancing the overall efficiency of AI interactions [23][24]. Group 5: AI Infrastructure and Security - Volcano Engine launched a multi-modal data lake solution to enhance data processing and analysis capabilities, significantly reducing data acquisition costs by 80% [26]. - AI security products were introduced, including AICC for secure computation and a large model application firewall to protect against attacks without increasing inference delays [30]. Conclusion - The AI product matrix unveiled by ByteDance at the conference illustrates a comprehensive strategy focused on model capabilities, toolchains, and infrastructure, aiming to enhance industry efficiency and innovation [31][32].

Core Viewpoint - ByteDance's Volcano Engine has launched the Doubao Model 1.6 series, which features multi-modal thinking capabilities and has achieved a score of 144 in the national college entrance examination mathematics. The model's unified pricing strategy reduces comprehensive costs by 63% for enterprises, leading to a significant increase in daily token usage [1][3][12]. Group 1: Model Performance and Market Position - The Doubao Model 1.6 has seen a 137-fold increase in daily token usage, reaching over 16.4 trillion tokens as of May 2025, capturing nearly half of China's large model market [2][4][7]. - The model is recognized for its advanced reasoning capabilities, ranking among the top globally in complex reasoning, competitive mathematics, multi-turn dialogue, and instruction following [3][20]. - Volcano Engine's revenue is projected to double to 25 billion, with token usage nearly equaling that of Alibaba and Baidu combined [4]. Group 2: Pricing Strategy - The Doubao Model 1.6 introduces a revolutionary unified pricing model, setting the cost for the most frequently used input range at 0.8 yuan per million tokens, resulting in a comprehensive cost of only 2.6 yuan [12][13]. - The pricing strategy is touted as the lowest in the industry, with a special discount zone further reducing costs for specific token requests [16][14]. Group 3: Technological Advancements - The Doubao Model 1.6 supports 256K context and features deep thinking, multi-modal understanding, and GUI operation capabilities, marking it as the first of its kind in China [20][21]. - The model's ability to interact with browsers and other software signifies a shift from passive response to proactive operation, enhancing automation possibilities for enterprises [21]. Group 4: Ecosystem Development - ByteDance is building a comprehensive AI cloud-native ecosystem, launching 12 tools for Agent development and enhancing its AI IDE product, which has over 1 million monthly active users [26][29]. - The company aims to transform traditional software development by positioning Agents as the new technological entity, capable of autonomous perception and task execution [28][31]. Group 5: Future Outlook - ByteDance's long-term strategy focuses on continuous investment in AI technology and infrastructure, aiming to establish a significant presence in the AI market and challenge existing paradigms [33].

在 FORCE 原动力大会上，火山引擎正式发布豆包大模型1.6、豆包·视频生成模型 Seedance 1.0 pro、豆 包·语音播客模型，豆包·实时语音模型在火山引擎全量上线，豆包大模型家族已成为拥有全模态、全尺 寸、高性价比的领先模型。截至2025年5月底，豆包大模型日均 tokens 使用量超过16.4万亿，较去年5 月刚发布时增长137倍。 智通财经APP获悉，6月11日，TikTok母公司字节跳动旗下云服务平台火山引擎举办火山引擎FORCE原 动力大会，公司总裁谭待正式发布豆包大模型1.6，并宣布将采用统一定价模式。价格方面，豆包大模 型再次降低，最低为2.6元/百万tokens，相较于豆包大模型1.5与DeepSeek-R1的7元/百万tokens的价格， 大幅下降63%。 在价格方面，豆包大模型1.6采用统一定价模式，首创按"输入长度"区间定价，在企业使用最多的输入 区间0-32K 范围内，豆包大模型1.6的价格为输入0.8元/百万 tokens、输出8元/百万 tokens，综合成本比 豆包1.5·深度思考模型、DeepSeek R1降低63%。Seedance 1.0 pro 模型每千 to ...

豆包大模型为企业 打通AI落地的"最后一公里"。 2025年，大模型在产业端加速行驶的轰鸣声正穿透技术的围墙。 这场始于技术、兴于场景的变革正在重塑各行各业的数字化版图。可以说，向产业化深入，就是当下人工智能最务实的命题。 当前，大模型正以前所未有的深度与广度融入千行百业，智能化转型的浪潮如火如荼：从金融业的个性化服务到营销场景的精准触达，到硬件终端的智能交 互到教育科研中的定制化学习。 企业纷纷投入资源拥抱AI，随着大模型不再是遥不可及的"奢侈品"，一个核心问题便也浮出了水面：是谁在支撑企业级AI落地的"最后一公里"？ 大模型产业化：向生态纵深 大模型热潮兴起以来，各行各业都在积极探索，如何接入大模型才能让其成为智能化转型的驱动力。 早在2024年，大模型应用就逐渐从早期探索阶段迈向了规模化落地。一个显著标志是，国内主流云服务商在2024年初"集体行动"，为企业提供大模型的一键 部署、深度集成能力等，大幅降低了大模型商用的门槛。 IDC发布的 《中国公有云大模型服务市场格局分析 1Q25》 报告显示，2024年中国公有云上大模型调用量达114.2万亿tokens。随着国产大模型从2025年初引 发全球关注， ...