Early in this month, reports surfaced indicating that DeepSeek is set to introduce its next-generation flagship AI model, DeepSeek V4, during the Lunar New Year festivities in mid-February this year. This new model is expected to boast enhanced code-writing capabilities. On January 20, which coincidentally marked the first anniversary of DeepSeek-R1's release, developers noticed that in the FlashMLA code updated by DeepSeek on GitHub, there were repeated references to an unidentified large model identifier, 'MODEL1.' This identifier is highly likely to represent the new model. 'MODEL1' stands apart from the existing 'V32' model in terms of its core technologies. The new architecture of 'MODEL1' is potentially tailored specifically for memory optimization and computational efficiency. Earlier, the DeepSeek research team had also published two technical papers, prompting the tech community to speculate that these research results might be integrated into the new model.