It is learned that at the 2025 Tencent AI Industry Application Summit held recently, Tencent (TCEHY) deeply discussed the technical details and design logic in the process of creating intelligent agent products, revealing Tencent’s layout in this emerging field.
Tencent pointed out that in the development of intelligent agent products, special attention is paid to the actual experience of users, and the question-answer function in the intelligent agent is strengthened. Enterprises can review and verify these question-answer pairs to ensure accuracy before publishing, thereby improving the reliability of intelligent agents in enterprise applications.
Google releases a new multimodal model
At the same time, Google (GOOG) recently held its annual I/O developer conference, bringing the expected massive AI application updates, and launched AI intelligent agents, Gemini 2.5 Pro deep thinking models, Veo 3 video models for the first time to achieve audio and video synchronization, search services are fully connected to AI, and the whole family bucket package is coming strongly-the technology giant is reshaping the industry landscape with disruptive innovation!
In terms of AI agents, Google is integrating intelligent agent capabilities into its full range of products, including Chrome browsers, search engines, and Gemini applications. Pichai said that Google’s agent tool Project Mariner can now supervise up to 10 tasks at the same time, including projects such as web searches.
In terms of real-time multimodal AI search, based on the AI review, Google officially launched an AI mode driven by the Gemini 2.5 Pro deep thinking model in the search engine. Due to the characteristics of AI, users can ask more complex and difficult questions to the search engine.
In addition, in the video area, the Veo 3 model not only generates higher quality and understands prompt words more accurately, but also generates video and audio simultaneously, including dialogues between different characters. Previously, Veo 2 will add tools such as camera control and object removal, and Imagen 4 can also accurately generate real text when generating pictures.
It can be said that in recent times, multimodal large language models have made significant progress in processing multiple input modalities such as images, audio, and video. Moreover, as an important product of the model layer in the AI industry, multimodal large models have benefited from policy support and have a good policy development environment.
WiMi highlights AI multimodal application orientation
According to the data, Wimi Hologram Cloud Inc (WIMI), a leading AI vision company, is vigorously developing AI and new generation information technology, accelerating the cultivation of a number of advanced and applicable artificial intelligence large model products, implementing infrastructure co-connection, industrial system co-construction, and innovation capability co-cultivation and other ecosystems, and effectively promoting the deep integration of large model technology and characteristic advantage industries, which will bring new development opportunities and momentum to multimodal AI large models.
So far, WiMi has focused on scenarios such as text-generated video and image-generated video, and supports cross-modal content creation. For example, the short video tool it developed can achieve low-threshold creation through image-generated video technology, and the technical framework is adapted to the open source tool chain. At the same time, WiMi reconstructs the content production process of film, television, advertising, etc. through multimodal technology, and cooperates with industries such as finance, automobiles, and education to customize solutions to improve corporate efficiency.
Conclusion
ChatGPT has been around for three years and has always attracted much attention since entering 2025. As the application scenarios of AI big models in the industry gradually become richer and the supporting industries become more complete, most companies have started to develop AI big models. In addition, the local industrial blueprints are planned in relevant documents to strongly support the development of AI models.
However, one thing that needs to be grasped is that for multimodal big models to move from technology to industrialization, it is necessary to form an “iron triangle” collaborative system between industry, academia and research. Through resource integration, scenario co-creation and standard co-construction, it is necessary to open up the “last mile” from technology research and development to large-scale application.
