On February 28th, Baidu released its financial reports for the fourth quarter and full year of 2023, showing a total revenue of 134.598 billion yuan and a net profit attributable to Baidu (non GAAP) of 28.7 billion yuan, a year-on-year growth rate of 39%; The fourth quarter revenue was 34.951 billion yuan, and the net profit attributable to Baidu (non GAAP) was 7.755 billion yuan, a year-on-year increase of 44%. In 2023, Baidu's annual revenue and profit exceeded market expectations.
AI has become a new driving force for Baidu's growth. The financial report shows that in 2023, Baidu's core revenue was 103.465 billion yuan, and the net profit attributable to Baidu's core (non GAAP) was 27.4 billion yuan, a year-on-year increase of 38%; In December 2023, the monthly active users of Baidu App reached 667 million. As of January 2, 2024, the total service volume of Carrot Run exceeded 5 million orders; In the fourth quarter, the number of Carrot Run service orders increased by 49% year-on-year to 839000, and the proportion of all autonomous driving orders in Wuhan reached 45%.
Robin Lee, founder, chairman and CEO of Baidu, said: "In 2023, we will make significant progress in iterating Wenxin's big model and ERNIE Bot's one word, restructuring products and services, and commercialization. At the same time, Baidu's core business will remain resilient and healthy. Looking ahead, we will continue to firmly invest in generative AI and basic models, laying the foundation for creating new growth engines."
Baidu CFO Luo Rong said, "Over the past year, we have continued to focus on improving operational efficiency; in 2024, we will continue to improve operational efficiency and achieve high-quality growth."
The call volume of Wenxin's large model ranks first in China
Collaborate with Chinese companies such as Samsung and Honor
In 2023, the usage of ERNIE Bot and the API calls of Wenxin's big model will maintain a rapid growth. Public data shows that by the end of December 2023, the number of ERNIE Bot users has exceeded 100 million; Robin Lee said in his speech in November last year that there are more than 200 major models in China, and "the call volume of Wenxin model is more than the total call volume of these 200 major models."
The Wenxin model has brought new growth opportunities for Baidu's performance. In the fourth quarter, top Chinese companies such as Samsung, Honor, and Autohome have all reached cooperation with Baidu. Taking Samsung as an example, its latest flagship phone, the Galaxy S24 series, integrates multiple capabilities of the Wenxin model, including calling, translation, intelligence, etc., especially the "circle and search" function, which has attracted attention. Users only need to make simple gestures such as drawing circles and lines on text and images to obtain accurate search results.
At the same time, the application of Wenxin large model technology and PaddlePaddle community ecology form a "flywheel effect". The financial report shows that PaddlePaddle has gathered 10.7 million developers, served 235000 enterprises and institutions, and created 860000 models based on PaddlePaddle. In PaddlePaddle Xinghe Community, there are more than 4000 innovative applications based on ERNIE Bot, covering many scenes such as marketing, education and training, and intelligent office.
AI refactoring unleashes commercial potential
Annual online marketing revenue of 75.1 billion yuan
In the past year, Baidu comprehensively promoted the AI proto biochemical reconstruction of its products and services, launched Baidu New Search, Baidu New Library, ERNIE Bot App and other AI native applications, as well as AI marketing tools such as Qingge and brand intelligence. According to the financial report, the monthly active users of Baidu App reached 667 million in December 2023, with online marketing revenue of 75.1 billion yuan for the whole year of 2023 and 19.2 billion yuan for the fourth quarter.
Many companies, such as Feihe, have started using AI native marketing tools based on the reconstruction of the Wenxin model. Taking the first brand of Chinese milk powder, Feihe, as an example, the "Feihe Official Intelligent Assistant" based on brand intelligent agents can provide users with more accurate and personalized services through multiple rounds of dialogue.
Baidu Wenku has been restructured into a "one-stop AI content creation platform", providing AI capabilities such as intelligent document, PPT, mind map, research report, and multi content generation and editing, as well as multi language document summary Q&A and merged reference generation. Combining natural language instructions for full scenario applications greatly improves creative efficiency and quality. According to public data, the number of Baidu Wenku AI users has exceeded 80 million, and the number of new features used has exceeded 700 million. According to a report from the National Center for Industry and Information Security, Baidu Wenku ranks first with 86 points in the field of PPT generation, and ranks first in multiple indicators such as generation quality and intention understanding.
The inference cost of the Wenxin big model has been reduced to 1%
Two major tools for intelligent cloud deployment to improve development efficiency
Thanks to the layered optimization of Baidu's unique four layer AI architecture, the inference cost of the Wenxin model continues to decrease, and training efficiency continues to improve. Robin Lee previously disclosed in the event that from March 2023 to the end of December 2023, Wenxin has basically reduced the reasoning cost to 1% of the original. At the same time, through the collaborative optimization of PaddlePaddle and Wenxin, the average weekly training efficiency of the model exceeded 98%.
In order to further improve the development efficiency of AI native applications and models, Baidu AI Cloud launched Qianfan AppBuilder and Qianfan ModelBuilder. As an AI native application development workbench, AppBuilder not only provides developers with rich development suites and application components, but also offers a lower threshold low code state. Users who do not understand code can quickly build their own AI native applications through natural language.
Qianfan ModelBuilder brings a complete set of large model development tool chains, supporting the development and fine-tuning of proprietary models. At present, Qianfan ModelBuilder has finely tuned 10000 models, providing developers with a complete large-scale model solution that combines proprietary models with basic large-scale models.