The Zhitong Finance App learned that Ping An Securities released a research report saying that with the penetration of large AI model applications and the continuous unlocking of more and more new scenarios, AI is gradually moving towards the B-side and edge side, and will also continuously require updates and iterations on hardware terminals and related chips. Under the trend of intelligence, the big model family upgrade has improved performance and lowered the threshold, and the ByteDoubao big model is expected to accelerate the landing on the AI side. On the B-side, the big bean bag model helps improve the production efficiency of enterprises; on the C-side, the end side application potential of the big bean bag model is prominent, and it is expected that the related industrial chain will usher in new opportunities.

Incident: On December 18, the ByteDance Volcano Engine Force Power Conference (Winter Conference) opened. Volcano Engine brought a full upgrade to the Doubao Big Model Family. In addition to releasing major models such as visual understanding, several large models released earlier were also upgraded.

Ping An Securities's main views are as follows:

The conference announced upgrades to the large model family to improve performance and lower the threshold to promote the implementation of large model applications.

Application is the most important way to monetize or monetize large models, and the market also expects large model manufacturers to bring good returns after large-scale capital expenditure. As one of the leading large models in China, the big bean bag model's development and implementation have also attracted market attention. At this Volcano Engine Power Conference, the company said that since the release of the Doubao Big Model, user usage has grown rapidly. The average daily token usage in December exceeded 4 times, an increase of more than 33 times over the May release period. The original power conference officially announced the release of a large visual understanding model and a 3D generated model, and also upgraded the Universal Model Pro, Wensheng Map model, music model, etc.

Specifically, the newly released visual understanding model has stronger content recognition and visual description capabilities, and is expected to be applied to scenarios such as education, e-commerce marketing, and tourism. The official price announced by the conference is 0.003 yuan/1,000 tokens. The price has entered the “cent” era. The lower usage cost will greatly reduce the usage threshold, which is conducive to the application and promotion of the big bean bag model.

The company has also made an important upgrade to the Universal Model Pro. As the big language model of the Doubao Big Model Family, the product capabilities released this time have been greatly improved. According to company data, the overall capability of the model is 32% higher than when it was first released, the professional ability is 54% higher than when it was first released, and the mathematical ability is 43% higher than when it was first released. Multiple indicators have already been aligned with GPT4O, and some complex scenarios have been surpassed, but the usage price of the Doubao Basic Model Pro is only 1/8 of GPT4O. In addition, the upgraded Wensheng Figure model implements practical functions such as one-click P-maps and one-click posters. The continuous optimization of the capabilities of the large bean bag model is empowering enterprises to increase their productivity.

On the B-side, the large bean bag model helps enterprises improve production efficiency

Due to the continuous improvement of capabilities, the Doubao Big Model has been used very effectively in key B-side fields, and the number of calls has also increased dramatically. According to data released by the company at this conference, the number of calls in the information processing field increased 39 times compared to the time of publication, the number of calls in the customer service and sales fields increased 16 times, and the hardware assistant and AI tool scenarios increased 13 times and 9 times, respectively.

Volcano Engine also stated at the conference that the multi-modal capabilities of the Doubao Big Model are being rapidly improved. Its voice input function uses its speed advantage to greatly improve input efficiency, and can also import various voice signals to recognize dialects and accents in different languages. At the same time, in order to promote the development of Doubao applications, Volcano Engine announced upgrades to the three platform products Volcano Ark, Button, and HiAgent to help enterprises better build their own AI competency centers and efficiently develop AI applications.

On the C-side, the end-side application potential of the large bean bag model is highlighted

In addition to being a tool to improve productivity, the big bean bag model is developing very actively on the end side, and the prospects for AI application implementation at the edge end are beginning to become clear. At this press conference, Volcano Engine said that the Doubao Big Model has cooperated with 80% of mainstream car brands. The AI cockpit scenario is one of the key directions of cooperation. Car users can interact with natural language, personalized services, and multi-modal interactions through the Doubao Big Model. On the mobile phone side, Volcano Engine is also actively incorporating intelligent assistants, system control, and creative tools into the domestic scene.

Volcano Engine shared an example of the Honor Mobile App Doubao Big Model at the conference, focusing on its abilities in magical retouching, etc. The company claims that Doubao has achieved remarkable results in terminal promotion. It has developed more than 50 application scenarios in the terminal field, covering about 0.3 billion terminal devices. The number of Dobao large model tokens calls from smart terminals has increased 100 times in half a year, and progress is very rapid.

Investment recommendations:

(1) In terms of AI computing power, we recommend Haiguang Information (688041.SH), IFF (601138.SH), Inspur Information (000977.SZ), Ziguang Co., Ltd. (000938.SZ), Zhongke Shuguang (603019.SH), Shenzhou Digital (000034.SZ), etc., and it is recommended to focus on Cambrian-U () and Jing Jiawei (300474.SZ); 688256.SH

(2) In terms of SOC, I recommend Hengxuan Technology (688608.SH), follow Lexin Technology (688018.SH), Xingchen Technology (301536.SZ), Zhongke Lanxun (688332.SH), Juxin Technology (), Quanzhi Technology (300458.SZ), etc.; 688049.SH

(3) In terms of memory chips, GigaYi Innovation (603986.SH) is recommended;

(4) For AI applications, we strongly recommend Zhongke Chuangda (300496.SZ), Jinshan Office (688111.SH), iFLYTEK (002230.SZ), Foxit Software (688095.SH), Wanxing Technology (300624.SZ), Desai Xiwei (002920.SZ), and recommend paying attention to Tonghuashun (300033.SZ) and Caixun (300634.SZ).

Risk warning: 1) Application implementation falls short of expectations; 2) Supply chain security risks; 3) Technological evolution and iteration fall short of expectations.