MemT-1.0: Towards Memory-Native Evolution

MemT-1.0 is a family of memory-native foundation models (Nano 36B-A3B and Flash 288B-A17B) that build memory into the model computation rather than relying on external retrieval: a dynamic local memory and a trainable hyper-memory consolidate past interactions into a fixed-size persistent memory state. At a 128K history length, Nano achieves about 1.86x end-to-end inference speedup and a 66x reduction in persistent storage compared with a full-context KV cache. Haotian Chen led the reinforcement-learning stage of the models.