Tencent s big model strategy panorama unveiled. One year ago, we proposed to build AI closest to the industry. At that time, companies were just beginning to explore and were concerned about whether they could be used and where they could be used. A year later, the question became more concrete: How do large models understand the business? How does AI fit into workflow? Today, at the 2025 Tencent Cloud AI Industry Application Summit , we gave a systematic solution - focusing on large model innovation, agent development, knowledge base construction and AI infrastructure upgrades to build a truly applicable AI native system to help more companies use AI, make good use of AI, and make AI a universal capability in the industry. At the same time, this is also the first panoramic appearance of Tencent's large model strategy: large model innovation: accelerated response, multi-modal interaction, and embracing open source. In 2025, Tencent's Hunyuan large model will take small steps and continue to improve its technical capabilities. During the summit, Tencent Cloud released the latest progress of the Hunyuan large model: Turbo S and T1 upgrades: On Chatbot Arena, a globally recognized and authoritative large language model evaluation platform, Hunyuan TurboS has climbed to the top eight in the world, second only to DeepSeek in China. Among them, Hunyuan TurboS also ranks among the top ten in the world for science abilities such as coding and mathematics. Enhanced multi-modal model capabilities: Hunyuan T1-Vision supports visual depth reasoning, Hunyuan Voice enables end-to-end voice calls, Image 2.0 can generate images in milliseconds in real time, and Hunyuan 3D has achieved a generational leap in controllability and ultra-high-definition generation capabilities; Multi-size models are fully open source: text, images, videos, and 3D modes have been fully open source, and Hugging Face, the Hunyuan 3D model, has been downloaded more than 1.6 million times. In the future, Hunyuan plans to launch multi-size hybrid inference models, ranging from 0.5B to 32B dense models, and activate 13B MoE models to adapt to the needs of enterprises and terminals. Multi-modal basic models such as hybrid images, videos, and 3D and supporting plug-in models will also continue to be open source. At present, the Hunyuan large model has been deeply applied in WeChat, QQ, Tencent Yuanbao, Tencent Conference, Tencent Documents and other products to help improve intelligence. As model capabilities evolve, Hunyuan will continue to export model capabilities through Tencent Cloud to help enterprises and developers innovate and improve efficiency. Agent development: From questioning and answering to working , building a large model of AI employees is just the brain . What can really do the work is the AI employee - the agent. During the summit, we officially upgraded the Tencent Cloud Large Model Knowledge Engine to an agent development platform , integrating multi-agent collaboration, workflow orchestration, and enterprise practical capabilities to create an agent system that truly can understand and execute. The Agent Development Platform has three core capabilities: Multi-agent collaboration: configure multiple Agents and handover logic with zero code, freely define collaboration strategies, and achieve efficient cooperation between agents. Flexible and controllable workflow execution: drag-and-drop to easily build business processes, innovatively introduce global agents, have a global perspective and node rollback capabilities, and fully upgrade the conversation experience. Enterprise practical capabilities: supports document comparison and efficient decision-making, accurate Q&A on large tables/large files; fully supports account-level, application-level, and knowledge base-level permission systems; and provides privatized deployment support for enterprise-level operation backends. At present, Tencent Cloud's intelligent agent platform has been implemented in multiple businesses inside and outside Tencent, helping users build their own intelligent agents. Knowledge base construction: Knowledge comes to life and AI understands the enterprise better. AI that is useful for enterprises requires enterprise knowledge . The real knowledge of an enterprise is hidden in thousands of documents, emails, rules and regulations, and employee experience. How to activate the knowledge accumulated in the enterprise? During the conference, Tencent Cloud released a new knowledge base product matrix: Tencent Enjoy: for enterprise organizations, upgraded to the Enjoy knowledge base, providing rich functions in knowledge integration and precipitation, knowledge update, permission management, AI question and answer, etc., to help enterprises truly integrate and utilize private domain knowledge and improve the efficiency of internal knowledge transfer within the enterprise; Tencent ima: Create an AI workbench for individuals and teams, integrating Hunyuan and DeepSeek large models to support building a personal knowledge base, initiating AI Q&A, quickly interpreting materials, and generating content in notes. The content generated by notes can also be added to the knowledge base. Creators can also publicly publish professional content through the Knowledge Account , and the system will correlate similar questions and answers raised by users in ima to improve the quality and influence of vertical questions and answers. During the main forum, Tencent Qidian officially released the marketing cloud intelligent agent. Based on capabilities such as large model reasoning, multi-agent architecture, marketing knowledge base, and MCP technology, marketing cloud agents automatically complete tasks such as crowd extraction, portrait analysis, product selection, marketing content production, and event review. Tencent Cloud Intelligent Computing series products have been comprehensively improved in three major directions: performance, reliability, and ease of use. It has released the first serverless platform in China that supports cross-card GPU scheduling; created a more cost-effective multi-machine interconnected network vRDMA; and combined with Xingmai network, AI server inspection system, cloud storage data Vientiane and GooseFS, vector database, intelligent log service, large model security and other products and solutions to provide solid infrastructure for large models and applications In addition, Tencent Cloud code assistant CodeBuddy also launched the Craft software development agent. Developers state their requirements in natural language, and Craft can automatically disassemble tasks, design modules, generate codes, and self-correct errors, which means that developers can develop applications in one sentence become a reality. We also shared the continuous evolution of products in tool chains, office collaboration, data analysis and other fields, including Tencent Conferences, Tencent Cloud BI intelligent assistant ChatBI, Tencent Questionnaires, Tencent Electronic Signatures, Tencent Documents, etc., further promoting AI to become a natural part of the daily tool chain of enterprises. Currently, more and more companies are moving from waiting and watching AI to applying AI or even relying on AI. The scenarios for implementation are expanding, the problems to be solved are deepening, and AI is changing from a tool to a capability, and from capability to productivity. In the future, we will work with more customers and partners to hone AI capabilities that are more usable and more trustworthy, and promote AI technology to enter thousands of industries and everyone s lives