2025 年 12 月 18 日，火山引擎在 2025 冬季 FORCE 原动力大会上发布了面向多模态 Agent 场景优化的豆包大模型 1.8 及音视频创作模型 Seedance 1.5 Pro，并同步更新了企业 Agent 落地支持体系与阶梯式成本方案。截至 12 月，豆包日均 token 使用量突破 50 万亿，同比增长超 10 倍，服务超百家万亿级企业客户，表明模型已在生产环境中实现规模化验证。Seedance 1.5 Pro 通过原生音视频联合生成架构实现毫秒级音画同步与多语言口型适配，其“Draft 样片”机制可提升约 65%创作效率。企业端，火山引擎推出 AgentKit/HiAgent 平台及“AI 节省计划”，以平台化工具与用量阶梯折扣推动 Agent 从试点走向规模化部署，构建“模型-平台-定价”一体化的竞争壁垒。

2025 年 12 月 18 日，火山引擎在 2025 冬季 FORCE 原动力大会上发布了面向多模态 Agent 场景优化的豆包大模型在 2025 FORCE 原动力大会开发者日（上海）上，火山引擎围绕智能体（Agent）规模化生产与应用落地的关键路径，对其开发者生态进行了系统化升级与集中发布。具体而言，在大模型服务平台火山方舟方面，公司推出新一代 Responses API、上线 Serverless RL（强化学习）平台，并增强了 VikingDB 向量数据库及 Viking 记忆库的功能；面向企业级应用，发布并升级了 AgentKit 企业级 AI Agent 平台底座能力，同时在开发工具侧推出扣子编程与 TRAE CN 企 业版。此外，公司亦将原有开发者社区升级为专注 Agent 的开发者社区，并同步上线动手实验室、核心开发者计划及城市社区等系列生态支持举措。

Insurance analytics involves the use of data analysis and statistical techniques to gain insights into the insurance industry. It helps insurers make informed decisions, assess risks, detect fraudulent activities, and enhance overall operational efficiency. This technology leverages data from various sources, including customer information, claims data, and market trends, to optimize underwriting, pricing, and claims processing activities.

The market by value in this report is defined as the revenues that enterprises gain from goods and/or services sold within the specified market and geography. • Values in this market are ‘factory gate’ values, that is the value of goods sold by the manufacturers or creators of the goods, whether to other entities (including downstream manufacturers, wholesalers, distributors and retailers) or directly to end customers. The value of goods in this market includes related services sold by the creators of the goods.

The global hybrid AI deployment market refers to the ecosystem of infrastructure, software, and services that enable organizations to develop, deploy, and manage AI models across a combination of on-premises data centers, private clouds, and public cloud environments. This strategic approach allows enterprises to process sensitive data locally for security, compliance, and low latency while leveraging the immense scalability and computational power of public clouds for intensive tasks such as training large language models. The primary applications span industries where data gravity and sovereignty are paramount, including financial services for fraud detection, healthcare for medical diagnostics, and manufacturing for predictive maintenance. Key innovations driving this market include container orchestration platforms like Kubernetes and unified management planes that provide a consistent operational experience across disparate environments. In the current landscape, hybrid AI is increasingly relevant as it offers a pragmatic solution for balancing cost, performance, and governance in enterprise AI strategies.

The market by value in this report is defined as the revenues that enterprises gain from goods and/or services sold within the specified market and geography. • Values in this market are ‘factory gate’ values, that is the value of goods sold by the manufacturers or creators of the goods, whether to other entities (including downstream manufacturers, wholesalers, distributors and retailers) or directly to end customers. The value of goods in this market includes related services sold by the creators of the goods.

The market by value in this report is defined as the revenues that enterprises gain from goods and/or services sold within the specified market and geography. • Values in this market are ‘factory gate’ values, that is the value of goods sold by the manufacturers or creators of the goods, whether to other entities (including downstream manufacturers, wholesalers, distributors and retailers) or directly to end customers. The value of goods in this market includes related services sold by the creators of the goods.

The global LLMs in education market encompasses the provision of artificial intelligence technologies, including software platforms, application programming interfaces, and integrated solutions designed to leverage the capabilities of LLMs (large language models) for educational purposes. These models function by utilizing deep learning algorithms trained on massive datasets to understand, generate, and manipulate human language. Primary applications span personalized tutoring systems that adapt to student progress, automated content creation for curricula, intelligent grading and feedback mechanisms, and sophisticated research assistants for higher education. Key advantages include the ability to deliver scalable, individualized learning experiences and to automate administrative workloads for educators. In the current market landscape, LLMs represent a transformative technology, poised to enhance pedagogical methods, improve learning outcomes, and drive operational efficiencies across K-12, university, and corporate training environments.

Digital biomarkers are measured across multiple layers of hardware (e.g., sensors) and software in medical devices that capture signals (behavioral and physiological data) from patients. Digital biomarkers, which collect trustworthy clinical data and enable ongoing monitoring and evaluation continuously and remotely, can improve the precision of diagnostics and treatment in healthcare.

Self-learning AI encompasses systems capable of independently enhancing their performance over time by learning from data and experiences without explicit programming. A core enabling technology is reinforcement learning, a machine learning paradigm where an intelligent agent learns to make optimal decisions through trial and error. This process involves the agent taking actions within a dynamic environment and receiving feedback in the form of rewards or penalties, allowing it to develop strategies that maximize a cumulative reward signal. Primary applications for this technology are transforming numerous sectors, including industrial automation, autonomous navigation in vehicles, algorithmic trading in financial industries, personalized treatment strategies in healthcare, and sophisticated natural language processing.

Predictive AI in retail market refers to the application of advanced analytical technologies, including machine learning, statistical algorithms, and data mining, to analyse historical and real time data. The primary objective is to forecast future events, consumer behaviours, and market trends with a high degree of accuracy. This technology enables retail organizations to transition from a reactive to a proactive operational stance by anticipating outcomes related to sales, inventory requirements, and customer preferences. By processing vast datasets that encompass transaction histories, customer interactions, and external market factors, predictive AI provides actionable insights. These insights empower retailers to make informed, data driven decisions across various functions such as merchandising, supply chain management, marketing, and customer service, ultimately aiming to enhance operational efficiency, minimize risks like overstocking or stockouts, and personalize the customer journey to foster loyalty and drive revenue growth.

Global predictive AI in stock market encompasses the application of advanced computational technologies, including machine learning, deep learning, and natural language processing, to analyze vast and complex financial datasets. This technology processes historical price data, corporate filings, macroeconomic indicators, and real time news sentiment to identify patterns and forecast future stock price movements, market trends, and volatility. The primary objective is to generate data driven, actionable insights that enhance algorithmic trading strategies, optimize portfolio management, and improve risk assessment frameworks. This market represents a significant shift from traditional technical and fundamental analysis toward more sophisticated, automated, and probabilistic models for investment decision making.

消费电子行业：行业热点聚焦，大厂陆续发布AI眼镜－产业链跟踪

数位转型进行式：TDX如何成为连结政策与企业的工具