Open source models allow anyone to use software for free for any purpose. DeepSeek, a rising Chinese AI startup, launched an open source AI model in January, causing a stir in the tech world. Third-party tests showed that the model matched or even beat rivals from OpenAI, Meta, and more. Notably, the company claims that the development costs were much lower.

deepseek bloomberg
DeepSeek's AI models are all open source. Photo: Bloomberg

“One thing I learned from DeepSeek is that the best open source models will have much broader adoption. When the model is open source, people will naturally want to try it out of curiosity,” Baidu CEO Robin Li said during an earnings call on February 18.

Last week, Baidu announced that it would open source its latest AI model, Ernie 4.5, starting June 30. The decision marks a departure from Li’s previous stance that large language models should not be open source.

He has argued that closed source models are more robust and economical than open source. In an interview with Yicai, he said that open source models are suitable for academia, not businesses.

Other tech leaders, including Yann LeCun, chief AI scientist at Meta, see DeepSeek’s success as a victory for the open-source model. According to LeCun, AI models should not be in the hands of a few people or companies. Open-source models benefit everyone.

Baidu also announced that it would make its chatbot free starting April 1 thanks to better technology and reduced costs. In late 2023, a few months after Ernie was launched, the company began charging 59.9 yuan for premium features.

Baidu faces competition from rivals such as Alibaba's Qwen and ByteDance's Doubao. Tencent said it would integrate DeepSeek into its super app WeChat.

(According to Insider)