RSI's three key puzzle pieces: MemT 1.0 fills in "memory" from the model layer

What happened: MemTensor open-sourced MemT 1.0, a memory-native model series that encodes experience as internal dynamic states read through a parallel memory branch in selected Transformer modules. Key numbers: MemT-1.0-Nano has 36B total parameters with 3B active, based on Qwen3.6-35B-A3B. MemT-1.0-Flash has 288B parameters with 17B active, based on DeepSeek-V4-Flash. Why it matters: The team frames memory-native design as a third scaling curve beyond inference, and positions memory as the first of three RSI building blocks, alongside feedback and verification/update mechanisms.

MemTensor open-sourced MemT 1.0, a memory-native model series that encodes experience as internal dynamic states read through a parallel memory branch in selected Transformer modules.

The team frames memory-native design as a third scaling curve beyond inference, and positions memory as the first of three RSI building blocks, alongside feedback and verification/update mechanisms.