Baidu BMU has added another strong support: it has brought in a mysterious researcher from North American Frontier Lab, who was formerly a core member of DeepSeek and is now in charge of Wenxin Multimodal.
Sun Tianxiang, head of Baidu’s Basic Model Research and Development Department (BMU), announced two personnel appointments. The first is the recruitment of Wu Qin, a senior researcher at the North American Frontier Lab, to strengthen the pre-training capabilities of Wenxin Large Model. The second is Wei Haoran, a former core member of DeepSeek, who has been appointed as the head of Wenxin’s multimodal algorithms. Wu Qin has participated in the development of large models at the North American Frontier Lab for several generations; his real name and background have not been disclosed. This adjustment is seen as Baidu’s effort to align itself with the pace of top international laboratories. Wei Haoran worked at DeepSeek from 2024 to 2025, participating in the development of V3.2 and V4 versions and leading the OCR series of open-source projects. After joining Baidu, he plans to open source the Unlimited OCR model on June 22, 2026, which has performed well in the OmniDocBench benchmark tests. Sun Tianxiang took over as head of BMU in July this year; he was previously one of the main researchers of MOSS large models at Fudan University.
Sun Tianxiang, head of the Baidu Basic Model Research Department (BMU), announced two personnel appointments: Wu Qin, a senior researcher at the North American Frontier Lab, will strengthen the pre-training capabilities of Wenxin’s large models; Wei Haoran, a former researcher at DeepSeek, will take on the role of head of Wenxin’s multimodal algorithms. Wu Qin has participated in the development of several large models at the North American Frontier Lab; his real name and background have not been disclosed. This adjustment aims to align with the pace of international top-tier laboratories. Wei Haoran worked at DeepSeek from 2024 to 2025, participating in the development of V3.2 and V4 versions and leading the OCR series of open-source projects. After joining Baidu, he led the team to open-source the Unlimited OCR model on June 22, 2026, achieving the world’s best performance in the OmniDocBench benchmark. Sun Tianxiang took over as head of BMU in July this year; he was one of the main researchers of MOSS large models at Fudan University. This adjustment is seen as Baidu’s effort to align its organizational structure and technical direction with…