It is reported that Baidu's Wenxin team has recruited two key talents in succession: a core author from North America's Frontier Lab and the core developer of DeepSeek OCR.

Beating AI Insight News Brief: Baidu's Basic Model Research and Development Department (BMU) is beefing up its core model team. Sources familiar with the matter said that department head Sun Tianxiang has brought in a researcher codenamed "Wu Qin" from a North American frontier lab to join the team and enhance Ernie's pre-training. The researcher has participated in the development of several generations of cutting-edge large models over the past two years, though their real name and former affiliation remain undisclosed. Separately, former DeepSeek researcher Wei Haoran and his team have also transferred internally to BMU, where he will serve as head of Ernie's multimodal algorithms. Wei Haoran's move had already been hinted at earlier. In June, when Baidu open-sourced Unlimited OCR, the technical director among the three core contributors was only credited under the pseudonym "YY", sparking community speculation that it was Wei Haoran. Wei Haoran is the first author of DeepSeek-OCR and DeepSeek-OCR 2; DeepSeek-V4's technical report also noted that he had left the company. When Unlimited OCR was released, it scored 93.92% on OmniDocBench v1.6, once setting a new record for end-to-end OCR performance. It adopts DeepSeek-OCR's DeepEncoder and redesigned the attention mechanism at the decoding end, enabling continuous parsing of dozens of pages of documents in one go. Sun Tianxiang took over BMU in July, simultaneously joining Baidu's Model Committee; at the end of the month, he was also admitted to Baidu's Technical Strategy Committee. Two months later, both of Ernie's core lines – pre-training and multimodal – have started adding personnel.