MemTensor introduces memory-native MemT-1.0 models

MemT-1.0 proposes fixed-size recurrent memory state as an alternative to growing context or retrieval; MemTensor claims 1.86× speedup and 66× less storage at 128K history, and reports SWE-bench gains. Hardware, VRAM fit, benchmark setup, independent reproduction, and the implications of its reported noncommercial code license remain unresolved.