On November 13, 2025, during the Baidu World Conference, Li Yanhong, the founder of Baidu, officially took the wraps off the Wenxin Large Model 5.0. This model represents a new generation of native all-modal large models. It leverages an autoregressive unified architecture, which enables seamless integrated understanding and generation. Additionally, it incorporates a hyper-sparse mixture-of-experts architecture, with a staggering parameter count that surpasses 2.4 trillion.

When subjected to testing, the Wenxin Large Model 5.0 showcased exceptional performance across various domains, including language processing, visual comprehension, audio interpretation, and visual generation. It is designed to support both multimodal input and output. Moreover, this innovative model has already been rolled out on Baidu's Qianfan platform. Users can simply log in to the platform to gain access to its powerful capabilities.