With the rapid development of AI technology, especially the frequent iteration of large models, developers are facing the challenge of existing applications quickly becoming outdated. However, this also provides developers with more choices and possibilities. By understanding technological trends and avoiding the extended line of development of large models, developers can create applications with practical value by leveraging the increasingly powerful model capabilities. These applications not only will not be overshadowed by the capabilities of large models but can penetrate more scenarios and enhance their value. For example, in the field of traffic safety event detection, the combination of cloud-based large models and edge system small models has significantly improved detection accuracy and reduced the workload of monitoring personnel. Such applications that can bring practical value are real opportunities for developers, and opportunities like these abound in various industries. Baidu has already generated significant effects by deploying large models and providing free call opportunities, as well as actively applying them in internal business lines.

Baidu Intelligent Cloud has released the Wenxin Large Model 4.5 Turbo and X One Turbo, aiming to solve current issues in AI models such as multimodal understanding, hallucination rate, cost, and speed. Wenxin 4.5 Turbo is a multimodal model that can integrate understanding of text, speech, images, and videos. Its API call price is only 1% of GPT4.5, and it performs excellently in the understanding of images and videos. X One Turbo enhances deep thinking and logical reasoning abilities, while also strengthening multimodal processing capabilities. It can call different tools at only half the price of similar models. The release of these two new versions marks the trend of multimodal models becoming the standard for future foundational models.

Discussed the price, performance, and multi-tool invocation capabilities of the Wenshen big model X One Turbo, demonstrating its advantages in AI application development through practical examples, and emphasizing the importance of cost reduction in driving AI innovation.

One important breakthrough in 2025 is the AI digital human technology, especially high convincing digital humans, which surpass the experience of real people through ultra-realistic voice and form, professional content, and flexible interaction. Such digital humans have huge application space in areas such as e-commerce, live broadcasting, and gaming. Compared to traditional digital humans, high convincing digital humans are driven by rich multi-mode scripts, not only providing vivid dialogues, but also adjusting expressions, tones, and movements in real-time, achieving smooth transitions between emotions and actions, with a performance far exceeding conventional standards, and even difficult to distinguish from real people.

In the field of e-commerce live streaming, digital human technology is being widely used to build efficient live streaming teams, achieving the effect of one person as a marketing team. Through the real-time scheduling of AI brains, digital humans can flexibly switch roles, such as hosts, co-hosts, etc., and promote sales based on the popularity and conversion rate of the live streaming room. In order to enable more people to have exclusive digital humans and monetize them, a one-click cloning function is introduced, which can train a reusable digital human for live streaming in just two minutes of video. Highly persuasive digital humans are typical applications of multimodal large models. Baidu Wenku has successfully created an AI function with over 40 million paying users by combining multiple models, demonstrating the practical effect of using multiple models in combination.

The liberalization department is able to handle a variety of modalities and file types of materials, including Word, PDF, images, audio, video, etc. Through AI search and image-to-image functions, users can efficiently research and create content. For example, when researching the impact of Yangtze River porpoises on the population of finless porpoises, users can drag and drop or paste various materials, specify the use of materials, and generate the required content with one click, such as long articles, PPTs, or video picture books. This ability is based on the combined use of multiple models, including Wenxin Jintiao model, multimodal model, proprietary model, and industry model, collectively forming the technical foundation of Cangzhou OS. Cangzhou OS includes check out plus and the two core components of the three libraries and three gases, which can vectorize and combine content from different modalities to generate new content. Based on this technical foundation, Baidu Wangpan recently launched the AI note function, further expanding the application scenarios of content generation and processing.

The product manager of Baidu Netdisk introduced the AI note function, which uses AI technology to automatically generate detailed notes with strong relevance to the original study materials. AI notes not only generate comprehensive knowledge content, but also provide timestamp tracking, knowledge tables, mind maps, and test questions based on video content, supporting multimodal content learning. Users can share notes, and this function has received positive feedback since its launch. Baidu Wenku and Netdisk plan to continue introducing more AI functions, aiming to become efficient learning and productivity tools.

Discussed the application of large visual models in the field of autonomous driving, emphasizing the safety and convenience of autonomous driving, as well as its global promotion trend. In addition, it also introduced the rapid development of intelligent agents in AI applications, especially the role of code agents in improving programming efficiency and no-code programming tools, such as Baidu's Wenxin Kuaima Commit and Miaoda, demonstrating how they help non-technical personnel easily complete tasks, as well as innovative applications in the fields of education and entertainment.

The instant platform helps college student entrepreneurial teams to quickly develop purchasing and delivery integrated systems, requiring only a few minutes and no need for a professional development team, significantly reducing the technical threshold. By uploading requirement documents, the instant intelligent body automatically completes tasks such as page design and function development, integrating Baidu's intelligent cloud storage, database, map navigation, and other practical functions. In addition, the instant platform also helps non-professional technical personnel such as Shaanxi fruit farmers, retired master craftsmen, and Suzhou embroiderers to realize their respective application dreams, demonstrating the future vision that ordinary people can also have programming capabilities after the popularization of technology.

Heartthinking App is a super versatile intelligent agent application that solves complex problems for users through autonomous planning and collaboration with multiple intelligent agents. It can provide detailed plans and solutions in various life scenarios such as travel planning, children's video picture book creation, and legal consultation. In addition, Heartthinking has also shown excellent performance in areas such as in-depth research, data analysis, and health consultation, integrating multiple tools and continuously optimizing the user experience. Heartthinking App has been launched on major Android stores and will release more beta features in the future.

Baidu Search Open Platform, through its AI Open Plan, aims to quickly connect users with various AI applications, including intelligent entities, H5 mini programs, and independent apps, providing users with the latest and most complete AI services, while also offering developers traffic and revenue opportunities. Through examples such as the integration of a 3D home design AI application, it demonstrates how users can quickly find and use AI services through search. In addition, Baidu Search will comprehensively index MCP servers in the market, providing developers with comprehensive development tools, emphasizing the importance of MCP in AI application development.

With the explosion of AI applications, developers are facing challenges such as tool standardization, platform adaptation, and component integration. The emergence of Model Context Protocol (MCP) is like installing a universal socket for AI, simplifying the development process and improving efficiency. Through MCP, developers can standardize their resource data capabilities output, while utilizing existing MCP server resources to reduce development workload and enhance application capabilities. Domestic and foreign technology giants such as Anthropic, OpenAI, Google, Alibaba, ByteDance, and Tencent are actively embracing MCP. Baidu is helping developers fully utilize MCP by optimizing the Wenxin large model, making the Qianfan platform compatible with MCP, and building the MCP server discovery platform, among other measures. Specific cases show the application of MCP in Samsung phones accessing Baidu Wenku NetDisk functions, intelligent bodies combining search, and MCP servers providing precise product recommendations and transaction services. Baidu will continue to promote the prosperity of the MCP ecosystem and empower developers to innovate.

Baidu has always provided developers with support such as models, development tools, and financial resources, promoting the prosperity of the large model ecosystem through the Wenxin Cup Entrepreneurship Competition. The two competitions have attracted over 2500 teams worldwide to sign up, and Baidu has provided over 200 million yuan in financial support and comprehensive assistance to the winning teams. Nearly half of the winning teams have received funding for the next round, showing a good trend of development. The third Wenxin Cup Entrepreneurship Competition has officially started, with increased support for entrepreneurs, the prize amount for single projects will double, and the investment amount for special prizes can reach up to 70 million yuan.

Baidu promises to increase the training of AI talents in the future five years, aiming to cultivate 10 million AI talents. At the same time, they introduced the first domestically developed 30,000-card cluster - Kunlun Core P800, which can effectively support large-scale AI model training, significantly increase chip utilization, and reduce energy consumption. In addition, Baidu also released a number of AI applications and technologies, including Wenxin large model, highly persuasive digital person, Cangzhou OS system, intelligent body instant progress, multi-agent collaboration APP, etc., aiming to provide developers with powerful tools and platforms to promote the development of the intelligent economy.

The conversation focused on the Baidu AI Developer Conference and provided a detailed introduction to artificial intelligence technology, especially the Wenxin large model's innovative applications in various fields. The discussion highlighted how AI technology can play a role in cultural relics protection and intangible cultural heritage martial arts inheritance. Through collaboration with the Wenxin large model, it is possible to achieve a deep understanding and cross-temporal dissemination of cultural relics, as well as systematic protection and inheritance of intangible cultural heritage martial arts. In addition, the conversation also introduced the application of code intelligent agents and intelligent code assistant Wenxin Kuaima in the field of software development, demonstrating how AI technology can lower the development threshold and improve development efficiency. Overall, it showcased the huge potential and actual effectiveness of AI technology in promoting cultural inheritance, enhancing work efficiency, and other aspects.

With the support of Baidu Intelligent Cloud, the Beijing Robotics Innovation Center has successfully developed the Tengong Satellite Robot. In a recent half marathon, the robot completed the race in 2 hours, 40 minutes, and 42 seconds, becoming the world's first humanoid robot to complete a half marathon. This achievement not only represents a breakthrough in sports events, but also serves as an extreme test of the long-term stable operation technology of humanoid robots in real-life scenarios. The center is leveraging the Tengong hardware platform and the general intelligent software platform to create a globally influential Jusheng Intelligent Innovation Base and application pilot industrial base. The Tengong robot is fully electrically driven, stands at 1.73 meters tall, weighs over 70 kilograms, and has 42 degrees of freedom. The upgraded Tengong 2.0 has enhanced artificial intelligence capabilities and upper limb operation abilities, enabling it to perform complex task planning and human-machine interaction. Additionally, the center has also released the world's first universal Jusheng Intelligent platform for one robot to assist multiple elderly people, which has already begun application trials in industrial, service, and educational scenarios.

The dialogue emphasized the key role of Baidu Intelligent Cloud in driving business innovation, especially in the efficient application development and model fine-tuning using large models. Through the Qianfan platform, enterprises can not only access and use a range of advanced models, such as self-developed Wenxin models and third-party models, but also effectively reduce costs and improve performance through model distillation technology. In addition, the Qianfan platform provides low-code and high-code development paradigms, as well as an enterprise-level agent framework, enabling enterprises to combine their own knowledge base and industry experience to customize exclusive intelligent bodies to meet specific business needs. Baidu Intelligent Cloud further promotes the openness of the ecology and the convenient use of data and tools by supporting the MCP protocol, encouraging developers to develop and share their own MCP services, jointly building a stronger ecosystem. Through the support of these technologies and platforms, enterprises like Minsheng Bank have successfully achieved industry innovation and practice.

China Minsheng Bank shared its progress and experience in the construction and application of artificial intelligence large models. The bank clearly stated that the large-scale application of AI technology is a strategic initiative, adopting a strategy of actively embracing, steadily advancing, and focusing on the development of systematic capabilities and implementing specific scenarios. Through close cooperation with partners such as Baidu, Minsheng Bank has built a large model technology support system, implemented more than forty applications, covering over one hundred and forty specific scenarios. It particularly emphasizes the importance of fine-tuning in model applications, as well as the decisive role of the completeness of the knowledge system in realizing the value of large models. In addition, Minsheng Bank is developing its first AI strategy, emphasizing the importance of talent, technology-driven, architectural transformation, and AI governance, aiming to achieve a leap from digitization to intelligence, while ensuring the security and ethical compliance of AI applications.

In this sharing session, two innovative products were introduced in detail - the intelligent customer service "Keyue" and the visual AI application "Yijian". The intelligent customer service "Keyue" achieves precise targeting of the target audience through a high emotional intelligence gold sales model, smoothly connecting to human customer service, as well as customer segmentation, old customer activation, and recovery functions. The demonstration showed how "Keyue" effectively helps customers solve problems with financial product refunds and recommends financial products that meet their needs, demonstrating its ability to understand and meet user needs. The visual AI application "Yijian" lowers the threshold for visual scene development through edge collaboration, enabling business personnel to participate in development and significantly reducing implementation costs. Through practical examples, it was demonstrated how "Yijian" can real-time detect whether hamburger production in the kitchen of a restaurant chain meets standards, effectively preventing customer complaints. These products not only demonstrate the application of intelligent infrastructure but also highlight their wide applicability and efficiency in various industries such as finance and catering.

The AI Developers Conference shared the complete process of AI application development, from selecting appropriate scenarios, utilizing the capabilities of large models to developing products, to paying attention to inference costs and model fine-tuning when scaling applications. Through cooperation cases with enterprises, such as the intelligent upgrade project of Chinese Steel Research, it demonstrated how to build the company's own AI infrastructure and promote the industry's intelligent transformation. The conference emphasized the importance of open systems, encouraged innovation, and further shared progress at the afternoon Developer's Ecosystem Conference, aiming to empower customers and partners comprehensively, so that artificial intelligence can truly serve to improve production efficiency and create a better life.