Chinese AI pioneer Baidu wants to make its Ernie 4.5 model open-source and free to use as the industry competition surges.
According to a Feb. 14 announcement, Baidu, a Chinese technology company that provides search engine services, cloud computing solutions, and artificial intelligence (AI) tools, will make its advanced Ernie AI model open source by the end of June.
A day earlier, the AI developer also notified that its chatbot Ernie Bot, upgraded with a deep search function, more powerful planning tools, and calling capabilities, will be available for free starting April 1 on both desktop and mobile platforms, after a year and a half of premium use only.
Open access invites contributions from a global community. Developers can modify, extend, and improve the model, accelerating progress and leading to new and unexpected applications. Moreover, making the code and architecture of a large language model (LLM) publicly available allows independent researchers and developers to scrutinize it for biases, vulnerabilities, or errors, thereby increasing accountability and public trust.
The move to an open-source model is part of Baidu’s fresh commitment to greater transparency in AI development. This strategy, already used by other major tech companies, not only promotes broader community engagement but also helps level the playing field in the rapidly evolving global AI landscape.
Nevertheless, before the move, Baidu CEO Robin Li had long advocated for closed-source models as the most viable path for AI development. Some of the perceived advantages of closed-source AI development are minimized risks of misuse or unauthorized modifications, security of proprietary algorithms and technologies, robust governance, and more chances to monetize the technology.
Therefore, the change in Baidu’s strategy may be an answer not just to the growing calls for openness in AI research and development but also to the growing competition in the field, particularly from the industry’s most-discussed newcomer – DeepSeek.
Launched only a few weeks ago, the DeepSeek open-source AI model is showing exemplary adoption. In its first week on the market, the platform’s traffic grew over 600%, while it attracted over 49 million visitors a day. In China, the AI model has already been integrated into the operations of several prominent domestic tech firms, particularly in the cloud services and semiconductor sectors. Besides, a few Chinese TV makers have reported incorporating DeepSeek’s highly popular AI assistant into their smart TV software.
Meanwhile, Baidu’s large language model Ernie, which allegedly matches GPT-4’s capabilities and debuted just a few months after OpenAI’s ChatGPT, has seen limited public adoption so far. Earlier this week, Robin Li acknowledged that Deepseek’s sudden emergence demonstrated the unprecedented nature of innovation. He also noted that the computing power constraints have driven Chinese companies to innovate to reduce costs.
Currently, there are about 200 large language models available in the Chinese market. The most popular are ByteDance’s Doubao, with 78.6 million active monthly users, DeepSeek – 33.7 million, and Ernie Bot with around 13 million users a month.
