PANews reported on March 17 that according to China Business News, on March 16, two years after the official release of Wenxin Yiyan, Baidu released the multimodal big model Wenxin 4.5 and Wenxin X1, which is comparable to DeepSeek. Wenxin Big Model 4.5 was launched on Baidu Smart Cloud Qianfan Big Model Platform, with an input price of 0.004 yuan/thousand tokens; Wenxin Big Model X1 input price is 0.002 yuan/thousand tokens, half of DeepSeek R1. In addition, Li Yanhong revealed in an internal speech this year that Baidu will release version 5.0 of the Wenxin Big Model in the second half of the year and increase the commercialization of AI applications. Next, Baidu will officially open source the Wenxin Big Model on June 30.
Through Baidu's native multimodal large model Wenxin 4.5, users can upload files including documents, pictures, audio, and video for AI to interpret. Wenxin Large Model X1 is a deep thinking model. The key technologies used include progressive reinforcement learning, end-to-end training based on thought chains and action chains, and a multi-unified reward system. Baidu claims that its performance is comparable to DeepSeek-R1 and has a "long thought chain." The reporter experienced that in addition to online search, multimodal capabilities and multi-tool calls have been added, which can understand and generate pictures, and call tools to generate code, charts, etc. At present, DeepSeek-R1 only supports text recognition for uploaded attachments.
