Wu Tian, vice-president of Baidu and deputy director of the National Engineering Research Center of Deep Learning Technology and Application, launches the latest version of the Wenxin AI chatbot. [Photo provided to chinadaily.com.cn]

Chinese tech giant Baidu released the latest version of its native artificial intelligence chatbot, Ernie Bot (Wenxin Yiyan) 5.0, on Thursday.

According to the company, the latest model boasts 2.4 trillion parameters and employs native full-modality unified modeling technology, enabling full-modality understanding and generation capabilities. It is designed to support the input and output of various information types, including text, images, audio and video.

Currently, individual users can download Ernie Bot through its official app and website, while enterprises and developers can utilize it through Baidu's Qianfan platform.

Wu Tian, vice-president of Baidu and deputy director of the National Engineering Research Center of Deep Learning Technology and Application, said that "unlike the majority of the industry's multi-modal solutions that adopt 'late fusion', the technical approach of Wenxin 5.0 utilizes a unified autoregressive architecture for native full-modal modeling.

"It jointly trains multi-source data such as text, images, videos and audio within the same model framework, enabling the full integration and collaborative optimization of multi-modal features under a unified architecture, thus achieving native full-modal unified understanding and generation."

Wu added that Wenxin 5.0 adopts a super-large-scale hybrid expert structure with ultra-sparse activation parameters, achieving an activation parameter ratio of less than 3 percent. While maintaining the model's powerful capabilities, it effectively enhances inference efficiency, she said.

According to Baidu, thanks to its native full-modality modeling, Wenxin 5.0 has achieved breakthroughs in multimodal understanding, coding and creative writing, enabling it to simulate the language style of the classic Dream of the Red Chamber novel. This has been described as generating a "Grand View Garden asset restructuring plan" that integrates classical context with modern business logic, demonstrating strong situational understanding and creative capabilities.

In the development of the latest Wenxin model, 835 experts from industries including technology, finance, culture, education, healthcare and energy, and disciplines such as mathematics, physics, chemistry, biology, literature, history and philosophy were employed. The experts provided guidance on knowledge impartation, appreciation and evaluation, and professional calibration, aiding the development of logical rigor, professional depth, creative quality and alignment of values.

Industry figures have said that the launch of Wenxin 5.0 signifies the maturation of the native full-modality technology approach, reflecting the underlying independent innovation capabilities of Chinese model manufacturers in multi-modality large models, which will further enhance China's technological influence on the global AI industry.