DeepSeek-R1 unveils new model "MODEL1" on its first anniversary

BlockBeats news, on January 21, according to Quantum Bit, DeepSeek-R1 revealed its new model "MODEL1" on the first anniversary of its release. DeepSeek updated the FlashMLA code on GitHub, where MODEL1 is mentioned in 28 places across 114 files, appearing as a different model from V32. It is known that V32 is DeepSeek-V3.2, so MODEL1 is likely a new architecture. The specific differences in the code are reflected in the KV cache layout, sparsity handling, and FP8 decoding, with several differences in memory optimization.