The Recurrent Looped Transformer (RLT) pairs a causal encoder with a recurrent decoder whose hidden state and sliding-window attention cache carry across all prompt and response tokens. In its concrete configuration, the model uses 48 encoder layers and 48 decoder layers, so the temporal path traverses 48t decoder blocks after t tokens. The paper notes that realized reasoning gains, hardware efficiency, and RL scaling remain to be established.
No score is assigned. Sources and their independence are shown in the citation chain below.