Chinese university unveils new AI model for agriculture

China Agricultural University (CAU) launched the Shennong Large Model 3.0 on Monday at the 2025 World Agrifood Innovation Conference (WAFI 2025) in Beijing, marking a significant move toward making AI more accessible and practical for agriculture.

According to Wang Yaojun, lead researcher at CAU, a major achievement is that this new model not only significantly reduces computational demands but also improves performance by 5 percent compared to the original version.

He added that the model's architecture was thoroughly redesigned to find the best balance between being lightweight and delivering high performance.

This latest release builds on essential groundwork from earlier versions. Shennong 1.0 established core agricultural knowledge and question-answering features, while Shennong 2.0 added multimodal functions and expanded the model's use in the agricultural sector.

The model is named after Shennong, a revered figure in Chinese mythology known as the "Divine Farmer," who is credited with teaching ancient Chinese people agricultural practices and the use of medicinal herbs.

Alongside the large model, the team also released an agent platform. This platform promotes a lightweight, deployable, and collaborative AI ecosystem that spans the entire agricultural supply chain, offering 36 specialized agents divided into six categories, including smart breeding, planting, and farming.

"These agents are designed for 36 distinct agricultural scenarios and can be integrated with agricultural machinery and sensors to enhance intelligent agricultural production," Wang said.

He noted that pilot programs have already been implemented in various areas around Beijing, as well as in north China's Inner Mongolia Autonomous Region and northeast China's Heilongjiang Province, offering localized services such as plant protection and customized guidance during the cultivation process.

The launch of Shennong Large Model 3.0 marks a new phase in agricultural AI – one focused on high efficiency, ease of use, and technological independence, thereby laying a strong foundation for the future of smart farming, Wang added.

Since its inception in 2023, the Shennong Large Model has been trained on a comprehensive and specialized agricultural dataset, including over 10 million agricultural knowledge graphs, 50 million records of modern agricultural production data, and 20,000 agricultural monographs.

WAFI 2025, taking place from October 12 to 14, has brought together around 780 experts from around the world, serving as a top-tier platform dedicated to advancing global agrifood innovation.