GuShiio.com 
Li Yanhong (Photo source: video screenshot)
The craze for DeepSeek has stimulatedBaidu, for many days in a row, the official announced major actions.
GuShiio.comAGI reported on February 14 thatJust now, Baidu announced open source. Baidu said it will launch the Wenxin Model 4.5 series in the next few months and will officially open source from June 30.
Earlier,Baidu Intelligent Cloud announced that it has successfully lit up the Kunlun Core third-generation 10,000-card cluster, and will light up the 30,000-card cluster in the near future. Even Baidu Intelligent Cloud, Wen Xiaoyan, etc. are connected to the DeepSeek model. Yesterday, Baidu announced that Wenxinyiyan will be available on April 1st. It will be fully free from 0:00 on the 1st of the month, and today it announced a major strategic adjustment for open source.
The debate between open source and closed source for large models has long been a hot topic in the industry. Baidu founder and CEO Robin Li previously insisted on the closed source route, saying that you should always choose closed source models. Open source models will become increasingly backward. Open source models themselves are an IQ tax.
Today, Robin Li and Baidu have changed their views on opening and closing sources. However, Baidu has not announced further open source information or the extent to which it will be open source in four months, which still takes time to verify.
It is reported thatOn February 7, 2023, Baidu officially announced the Wenxinyiyan project, indicating that it is a new version of a generative AI product equipped with Baidu Wenxinyan big model technology; then on March 16, Yanhong Li officially announced the launch of Baidu big language model Wenxinyiyan invitation test to demonstrate its capabilities in literary creation, commercial copy creation, etc., becoming the first AI chat robot product in China released after the ChatGPT craze.
In fact, as early as January 2023, Baidu internally tested the relevant technical functions of Wenxinyiyan in the search and library business. As of June 26, the Wenxinyan Model will be upgraded to version 3.5, with a comprehensive upgrade in terms of creation, question and answer, reasoning and coding capabilities, and a new plug-in mechanism; August 31, 2023: Wenxinyiyan will be fully open to the entire society, and users can use it on the Wenxinyiyan APP and official website.
In November 2023, Baidu’s Wenxinyiyan Professional Edition based on Wenxinwen Model 4.0 was officially launched, and membership services were launched. The monthly price is 49.9 yuan/month. Members can use China’s model Wenxinyiyan 4.0 released in October 2023 and is still Baidu’s latest model. In addition, it also has rights such as image generation and web plug-ins, making it the first large-scale model product in China to use the membership model to charge C-terminal charges; In June 2024, Wenxin Model 4.0 Turbo was released; on September 4, 2024, Baidu officially announced Wenxin Yiyan APP was officially upgraded to“ Wen Xiaoyan”.
Now, with China’s open source model DeepSeek attracting attention and impacting the basics of chat robot payment and closed-source technology, Baidu has begun to readjust its business direction.
On February 3, Baidu Intelligent Cloud announced that DeepSeek-R1 and DeepSeek-V3 models have been put on the Baidu Intelligent Cloud Qianfan platform, simultaneously launching ultra-low price plans and providing free services for a limited time of two weeks; On February 5, Baidu Intelligent Cloud successfully lit up the Kunlun Core third-generation million card cluster, and will further light up the 30,000 card cluster; on February 8, Baidu Intelligent Cloud’s four major model application products, Keyue, Xiling, Yijian, and Zhenzhi, were officially launched to access a new version of the DeepSeek model.
On February 12, according to CNBC, Baidu announced that it plans to release the next-generation AI model Ernie 5.0 in the second half of this year, which will significantly enhance its multimodal capabilities. At the same time, The information reported that Apple is still cooperating with Baidu to develop the iPhone’s national AI function.
Public data shows that as of November 2024, the number of Wenxinyiyan users is 430 million, and the average daily calls of Wenxinyi Model exceed 1.5 billion, an increase of more than 30 times compared with 2023.
On February 11, local time, at the World Governments Summit held in Dubai, when asked if he was surprised by DeepSeek’s progress, Robin Li did not respond directly, only saying that innovation could not be planned.
Li Yanhong also said that the essence of innovation is to continuously reduce costs and thereby improve productivity.
When asked whether large models can be highly invested and commercialized, Li Yanhong believes that technology companies need to maintain their technological leadership, so they need to expand investment. He is optimistic about to-B (enterprise-oriented) applications, and points out that there is still a significant gap between to-C (consumer-oriented) applications and super applications in the mobile Internet era. Currently, there is no way for big models to become super applications.
When talking about opening and closing sources of models, Li Yanhong said that his feeling in the past few months is that open source models can help spread large models, and wider dissemination means wider use.
“What I’ve learned over the past few months is that open source can help you get more attention. We are in the early stages of AI and generating AI innovation, and faster dissemination will help increase adoption rates, but it will also help more people try the technology, thereby enabling innovation at the application level.” rdquo; Li Yanhong said.
This means that Robin Li has changed his attitude towards the open source model, and Baidu is constantly reflecting and adjusting internally.
Recently, OpenAI CEO Altman also said that OpenAI is on the wrong side of history and the company is considering adjusting its open source strategy.
In fact, there is no need to laugh at or ridicule Baidu’s drastic adjustment of its big model strategy. This instead proves the rapid development and changes of the AI big model industry, and the entire industry is still in a very early stage.
