Last Thursday, Tencent announced the official launch of the app "Tencent Yuanbao", which is based on the hybrid model. Previously, Tencent also created a hybrid assistant for the C-side, but only the applet and the PC side, and the ingots are the first AIGC pure application created by Tencent for the C-side.

Compared with the C-end applications of other large manufacturers, what is special about "Tencent Yuanbao"? With the release of application products from major technology giants, what will be the next step for large-scale model applications?

AI search is the core application scenario

Compared with Baidu, Alibaba, and ByteDance, which launched related products last year, Tencent's move has been a year late. In this regard, Liu Yuhong, vice president of Tencent Cloud and head of Tencent's hybrid model, explained again at the Tencent Yuanbao launch conference on May 30, "Tencent is not the first to make a big model. "It's still a very early stage, even if it's three months or half a year late, it's not too important in the long run."

Similar to competitors such as Wenxin Yiyan, Tongyi, Doubao, Kimi Intelligent Assistant, Zhipu Qingyan, Conch AI, etc., Tencent Yuanbao App also provides a series of functions such as AI search, AI mapping, dialogue, document summary, AI writing, etc., and the overall user experience is only slightly different.

Tencent Yuanbao's product slogan is set as "easy work, more life". This slogan shows that Tencent Yuanbao's goal is to make both efficiency tools and content products, which are both easy to use and fun. Compared with the industry, ByteDance and Minimax currently use a multi-App strategy, scattering different application scenarios in different Apps; Zhipu AI, Baidu, and Alibaba prefer to integrate all scenarios on the same app.

Tencent's positioning of the Yuanbao App as an all-round AI assistant. From the perspective of the product interface, AI search is still its core application scenario.

According to the statistics given by Tencent Yuanbao, the penetration rate of large model applications in the whole industry is still very low, mainly due to the fact that "large models are still in the development stage, and AI product solutions are not mature" and "user awareness is still being established, and the demand for AI is focused on search/efficiency".

The application of large models is in full bloom

Behind the upgrade of Tencent's ingot's product capabilities is the continuous iteration of the underlying model of Mixed Yuan. Since its debut in September 2023, the parameter size of Tencent's hybrid model has been upgraded from 100 billion to trillions, and the scale of pre-trained corpus has been upgraded from trillion to 7 trillion tokens.

Liu Yuhong said, "In the past year, we have continued to promote the ability of Tencent's hybrid model, polishing technology in rich and massive business scenarios, and at the same time gaining insight into the real needs of users, hoping that Tencent Yuanbao can become a good partner and helper in the lives of users, and ultimately serve the lives of every ordinary person." ”

Today, the hybrid model has "blossomed" inside and outside Tencent. In the first quarter of this year, Tencent collaboration SaaS products such as WeCom, Tencent Meeting, Tencent Docs, Tencent Lexiang, Tencent e-Sign, and Tencent Questionnaire have all been connected to Tencent's hybrid model to achieve intelligent upgrades. Tencent Hybrid has also made continuous efforts to help the intelligent upgrading of the industry, and has been applied to many industries such as automobile, retail, education, finance and healthcare, bringing new growth points to the enterprise service business. In April this year, Tencent released the "Global Intelligence" solution for the automotive industry, working with more than 10 car companies to start industry model cooperation in multiple core scenarios such as R&D, production, and marketing.

The price of Tencent's hybrid model has been reduced across the board, and one of the main models, the hybrid element-lite model, the total length of API input and output is planned to be upgraded from the current 4k to 256k, and the price has been adjusted from 0.008 yuan/1000 tokens to be fully free.

There is no rush to commercialization

The commercialization of AI large models has always attracted much attention, but commercialization is not the most important thing for Yuanbao App now.

When asked about this question on the same day, Liu Yuhong said that it is indeed too early to talk about the commercialization of the entire large model industry, and it is still in the stage of exploration. At the same time, he mentioned that Tencent will open up the capabilities of the hybrid model to Tencent Meeting, advertising, Tencent Cloud and other products, which have their own commercialization paths, "but for the Yuanbao App itself, we have no consideration in this regard for the time being." ”

It should be mentioned that in the B-end market of large models with a higher degree of commercialization, after Byte's bean bag large model fired the first shot of price reduction, large model manufacturers set off a fierce price war in May and ran into the free era. When communicating with reporters, many industry insiders believe that the purpose of the price war of large model manufacturers is more to cultivate market awareness and accumulate strength to stake land.

At present, everyone is exploring the path of commercialization, because the industry is still in the early stage of development, it is still a large input and low output, and it is necessary to burn a lot of money to further improve technological innovation and improve the maturity of AI.

The large-scale model application market is still in a period of rapid development, and with more players of AI products at home and abroad "participating in the war", the consumer market will usher in more intelligent and efficient products and services, and the competition of large-scale model apps may enter a more fierce new stage in the future.