BlockBeats news, January 21st, according to QuantumBit reports, on the first anniversary of DeepSeek-R1’s release, the new model “MODEL1” has been unveiled. DeepSeek updated the FlashMLA code on GitHub, with MODEL1 mentioned 28 times across 114 files, appearing as a distinct model from V32. It is known that V32 is DeepSeek-V3.2, and MODEL1 is likely a new architecture. Specific differences in the code are reflected in KV cache layout, sparsity handling, and FP8 decoding, with multiple variations in memory optimization.
