On the evening of November 21st, Baidu released its third-quarter financial report for 2024, showing a quarterly total revenue of 33.6 billion yuan, Baidu's core revenue of 26.5 billion yuan, and a 17% year-on-year increase in net income attributable to Baidu's core to 7.54 billion yuan, exceeding market expectations. During the evening earnings call, Baidu's founder, director, and CEO, Robin Li, introduced that in November, the Wenshen large model's daily invocation volume had reached 1.5 billion and revealed that Baidu will release a new version of the Wenshen large model in early 2025.
In the third quarter, the intelligent cloud business continued to grow rapidly, with revenue increasing by 11% to 4.9 billion yuan, and AI revenue accounting for over 11%. Baidu is continuing to advance the restructuring of search with large models. Currently, generative AI content on the Baidu App covers nearly 70% of monthly active users; over 20% of search results pages contain AI-generated content.
During the call, Robin Li stated that Baidu is committed to a global strategy centered around artificial intelligence. Enhancing AI capabilities is a long-term strategic focus for Baidu and is the cornerstone of its transformation driven by AI. He pointed out that this strategy will enable Baidu to seize significant long-term growth opportunities.
The new version of the large model will be released in early 2025.
Robin Li stated that thanks to the enhanced capabilities, wide applications, and improved cost efficiency of the Wenshen large model, its daily invocation volume has significantly increased, reaching 1.5 billion times in November, processing over 1.7 trillion text tokens daily; compared to the initially disclosed 50 million daily invocation volume a year ago, it has grown 30 times. He revealed that Baidu will launch a new version of the Wenshen large model in early 2025 to consolidate its leading edge on the base model.
During the quarter, Baidu continued to iterate on the capabilities of the Wenshen large model, optimizing model efficiency. Based on its independently developed four-tier AI technology stack optimization, the flagship model, Wenshen 4.0 Turbo, has significantly improved its inference efficiency, with a 48% performance improvement since its release in June.
At the same time, Baidu 'customizes' basic models for different application scenarios, expanding the portfolio of lightweight model products. In the third quarter, Baidu launched two enhanced lightweight models, ERNIE Speed Pro and ERNIE Lite Pro, compared to the previous quarter's ERNIE Speed and ERNIE Lite, the enhanced models have lower latency, higher throughput, better stability, and higher accuracy.
In the multimodal field of image generation, Baidu recently released the retrieval-enhanced document and image technology 'Wenxin iRAG' (image based RAG), extending RAG technology from text generation to image generation, effectively reducing the hallucination problems of document and image, greatly improving the accuracy of AI document and image, applicable to various scenarios such as film and television works, comic works, and poster production.
Intelligent Cloud revenue of 4.9 billion yuan, AI revenue proportion increased to over 11%
Driven by generative AI and basic model advancements, China's cloud computing industry is accelerating its transition to intelligent computing. In the third quarter, Baidu's Intelligent Cloud revenue reached 4.9 billion yuan, an 11% year-on-year growth, with Non-GAAP operating profit margin improvement. The revenue contribution of AI increased to over 11%, compared to last year's fourth quarter when this figure was first disclosed, it was only 5%.
Baidu's technological advantage in AI infrastructure is a key factor driving the growth of the Intelligent Cloud business. This quarter, Baidu Intelligent Cloud further enhanced the management capabilities of tens of thousands of card clusters, achieving 99.5% effective training time. At the same time, through ModelBuilder, lightweight models can be fine-tuned into proprietary models, with performance comparable to Wenxin 4.0, lower cost, and faster inference speed.
During the earnings call, Baidu Intelligent Cloud's growth was mainly driven by high demand for model training and inference from industries such as internet, education, and finance, with incremental revenue from mid-sized enterprise customers growing by 170% month-on-month. According to the IDC report, in the Chinese public cloud AI market, Baidu Intelligent Cloud has been the leader for five consecutive years.
This quarter, Baidu established partnerships with enterprises such as Shansong, Kingdee, and Tuhu Car Care. In the dining sector, Yum China deployed the AI smart customer service system 'Keyue' based on the Wenxin large model, with Keyue API daily call volume reaching millions at peak times; in the education sector, Exam Treasure achieved intelligent analysis of test questions by calling the Wenxin large model, reducing the cost of analyzing a single test question by 99.8%, the proportion of paying users increased by over 100%, and revenue increased by 246%.
Baidu App's AI features have covered nearly 70% of monthly active users
With the continuous restructuring of baidu's search by ai, currently, baidu's app ai function has covered nearly 70% of monthly active users. Users who interact with ai have longer usage times, with total daily interactions reaching tens of millions. According to the introduction, the ai functions of search can provide rich content such as ai summaries, images, videos, smart bodies, digital humans, etc., enhancing user experience.
Smart bodies are an important form enriching baidu search content and experience. Currently, the Wenxin Smart Body Platform has gathered 150,000 companies and 800,000 developers, covering various scenes such as companies, tools, workplaces, roles, entertainment, etc. Taking the BYD official website's smart body as an example, after its launch, its sales lead conversion rate increased by 119%.
Another internal product, baidu Wenku, is undergoing transformation through ai restructuring. In September of this year, baidu Wenku's ai function's monthly active users (MAU) exceeded 50 million, a year-on-year growth of over 300%; and user willingness to pay continues to strengthen, with third-quarter subscription revenue increasing by 23% year-on-year. At the recent Baidu World 2024 event, Baidu Wenku and Baidu Wangpan jointly created the tool-type smart body 'Free Canvas,' attracting over 200,000 people to queue for reservations on the first day.
At the same time, baidu is further lowering the threshold for ai application development to promote the ecological prosperity of ai applications. Recently, baidu released a no-code development tool called 'Miaoda', supporting users to develop applications in natural language. After the announcement, Miaoda quickly attracted over 5000 companies queuing to test. Li Yanhong stated, 'We believe that expanding the capabilities of artificial intelligence beyond the programmer population will drive the next wave of innovation, unleashing tremendous social value.'
