The Basic Model Research Department (BMU) of Baidu has recently made significant talent arrangements, continuously accelerating the comprehensive upgrade of organizational structure and core technologies. According to the information, the team has successfully recruited experienced researcher Wu Qin from a renowned Frontier Lab in North America, to comprehensively strengthen the pre-training foundation capabilities of the ERNIE large model. Meanwhile, Wei Haoran, a former researcher at DeepSeek, has officially transferred to BMU and is now serving as the head of the ERNIE multimodal algorithm team.
This key team expansion comes at a crucial time when Sun Tianxiang, who took over as the head of BMU in July this year, is actively pushing forward the upgrade of the ERNIE large model. Through dual innovations in organizational structure and cutting-edge technology, Baidu is trying to further keep up with and align with the research rhythm of top international artificial intelligence laboratories.
Looking back at the recent technical iteration roadmap, Baidu has consistently maintained a high frequency and intensive update pace in the field of foundational models. As early as the beginning of 2025, Baidu Search announced full integration of the deep search functions of DeepSeek and ERNIE large model. Subsequently, the ERNIE large model 4.5 series was officially announced to be open-sourced at the end of June of the same year, and an X1.1 upgrade version was released in September, significantly improving the factual accuracy of content by 34.8%. Entering November 2025, Baidu officially launched the ERNIE large model 5.0 version, which broke through 2.4 trillion parameters, marking a native full-modal model. After entering 2026, with the successive joining of several high-level external and industry core researchers, the ERNIE large model has once again taken a new step forward in organizational efficiency and technological breakthroughs.