Less is More: Recursive Reasoning with Tiny Networks
This work addresses the challenge of efficient reasoning for AI systems, offering a domain-specific incremental improvement in puzzle-solving tasks.
The paper tackles the problem of solving hard puzzle tasks like Sudoku, Maze, and ARC-AGI with small neural networks, achieving 45% test-accuracy on ARC-AGI-1 and 8% on ARC-AGI-2 using a tiny recursive model with only 7M parameters, outperforming large language models with far fewer parameters.
Hierarchical Reasoning Model (HRM) is a novel approach using two small neural networks recursing at different frequencies. This biologically inspired method beats Large Language models (LLMs) on hard puzzle tasks such as Sudoku, Maze, and ARC-AGI while trained with small models (27M parameters) on small data (around 1000 examples). HRM holds great promise for solving hard problems with small networks, but it is not yet well understood and may be suboptimal. We propose Tiny Recursive Model (TRM), a much simpler recursive reasoning approach that achieves significantly higher generalization than HRM, while using a single tiny network with only 2 layers. With only 7M parameters, TRM obtains 45% test-accuracy on ARC-AGI-1 and 8% on ARC-AGI-2, higher than most LLMs (e.g., Deepseek R1, o3-mini, Gemini 2.5 Pro) with less than 0.01% of the parameters.