The Kimi K3 model, developed by China's AI company Moonshot AI, has recently become embroiled in yet another controversy surrounding model distillation. According to an investigation conducted by OpenAI, a coordinated effort to extract confidential reasoning content from its model has been traced back to individuals affiliated with Moonshot AI. This campaign reportedly generated approximately 16,000 requests per day, involving more than 15,000 users. In response, Huang Zhenxin, the head of enterprise business at Moonshot AI, clarified that the significant performance leap of the K3 model is attributable to its innovative underlying architectures. These include the KDA hybrid linear attention mechanism, Attention Residuals technology, and the Moon Clip second-order optimizer, rather than any form of model distillation or replication. He underscored that the training of the K3 model relies solely on domestically produced computing clusters and original technologies, with all data sourced from publicly compliant channels. Despite the ongoing controversy, the K3 model has emerged as one of the world's largest open-source models, boasting 2.8 trillion parameters, and has claimed the top spot on the Frontend Code Arena leaderboard, demonstrating the technical prowess and innovative capabilities of Chinese AI enterprises.