Based on insights from individuals close to the situation, DeepSeek convened a confidential meeting this past Sunday, inviting investors to its offices in either Beijing or Hangzhou. The company's CEO, Liang Wenfeng, joined the session remotely. To maintain confidentiality and prevent any potential information leaks, investors were asked to hand over their electronic gadgets and personal belongings upon arrival, permitting only the use of paper and pen for taking notes. During the course of the meeting, Liang Wenfeng disclosed that DeepSeek is presently in the process of training a model boasting 2 trillion parameters. This new model surpasses the size of their current flagship V4 model, which has 1.4 trillion parameters. Additionally, he outlined the company's ambitious plans to develop an even larger model with 8 trillion parameters, signaling DeepSeek's active engagement in the competitive arena of large-scale model development.