PANews reported on January 21 that, according to QuantumBit, the name "MODEL1" appeared for the first time in DeepSeek's updated FlashMLA code on GitHub, appearing in 28 mentions across 114 files and listed alongside the existing version V32 (DeepSeek-V3.2), suggesting that MODEL1 is a next-generation architecture model. Code differences indicate that the model has been optimized in areas such as KV cache layout, sparsity handling, and FP8 decoding, and may be officially released around the Spring Festival. Combined with the recently disclosed mHC residual connection mechanism and Engram memory module, MODEL1 is expected to integrate several self-developed innovations.