Baidu, the leading technology firm from China, is set to make its Ernie artificial intelligence (AI) model available as open-source starting Monday. This initiative was first announced in February, coinciding with the company offering its chatbot to the public free of charge. Industry observers are interpreting this move as a pivotal shift in China’s AI landscape, marking the first occasion a prominent AI developer has opted to share its proprietary technology with the global community. Analysts suggest that Baidu’s action could have ramifications similar to those seen when DeepSeek-R1 was originally introduced.
Baidu to Gradually Open-Source Ernie
On February 14, Baidu disclosed plans to release its flagship large language model (LLM) Ernie to the open-source community. A post on WeChat indicated that the company would be making its “next generation of AI model” accessible; this is widely believed to refer to Ernie 5. In March, Baidu had previously launched the Ernie 4.5 foundation model, alongside its inaugural reasoning model designated as Ernie X1.
While the company is transitioning towards an open-source framework, the extent of this release remains unclear. It is yet to be determined whether the AI models will be entirely open-source or only partially accessible. Full open-source status implies that the company releases the model weights, training methodologies, and architectural details, enabling developers to freely use, modify, replicate, and build upon the models.
Conversely, a partial open-source approach would mean that only the model weights are shared, keeping further technical details confidential. In this scenario, developers would be limited to using and fine-tuning the model without the ability to replicate it. A fully open-source example is Mistral NeMo 12B, while Meta’s Llama 3 represents a model with only open weights.
Experts contend that Baidu’s decision could significantly contribute to the democratization of AI worldwide. Sean Ren, an associate professor of computer science at the University of Southern California and recognized as Samsung’s AI Researcher of the Year, noted to CNBC, “This isn’t just a China story. Every time a major lab open-sources a powerful model, it raises the bar for the entire industry.”
Interestingly, the last occurrence of a Chinese open-source AI model (DeepSeek-R1) had a detrimental effect on the stock prices of various leading AI companies, resulting in a net negative impact on several US-based enterprises.
