Hierarchical Reasoning Model

A revolutionary 27M-parameter AI model that performs complex sequential reasoning in a single forward pass. Featuring dual recurrent modules for high level planning and sharp detail, it outperforms larger models on puzzles and maze challenges.

The Hierarchical Reasoning Model (HRM) is a brain-inspired AI system designed for efficient, multi-level reasoning and planning. Unlike traditional large language models that rely on resource-intensive Chain-of-Thought techniques, HRM performs complex sequential reasoning in a single forward pass. It features dual recurrent modules—one for abstract high-level planning and another for rapid, detailed computations—enabling stable and efficient training. Despite its compact size (27M parameters), HRM achieves near-perfect performance on challenging tasks like Sudoku puzzles and maze navigation, using only 1,000 training samples without pre-training. It even outperforms larger models on the Abstraction and Reasoning Corpus (ARC), a key benchmark for artificial general intelligence. HRM offers a promising step toward universal reasoning systems with minimal computational overhead.