This August 2025 paper introduces the Hierarchical Reasoning Model (HRM), a novel AI architecture inspired by the human brain's hierarchical and multi-timescale processing. This model aims to overcome the limitations of current large language models (LLMs) and Chain-of-Thought (CoT) techniques in complex reasoning tasks, which often suffer from computational inefficiencies and extensive data requirements. HRM utilizes two interdependent recurrent modules: a high-level module for abstract planning and a low-level module for detailed computations, enabling it to achieve significant computational depth. Notably, HRM demonstrates exceptional performance on challenging reasoning benchmarks like Sudoku and maze navigation with minimal training data (around 1000 samples) and without pre-training or CoT data. The research further explores the brain-like hierarchical dimensionality organization within HRM, where the high-level module operates in a higher-dimensional space, mirroring principles observed in the mammalian cortex.
Source:
https://arxiv.org/pdf/2506.21734
