Paper summary: Less is More: Recursive Reasoning with Tiny Networks
Tiny Recursive Model, a single 2-layer network with 7M parameters that repeatedly refines its answer, reports 45% on ARC-AGI-1 and 8% on ARC-AGI-2, above many large LLMs on these puzzles.
Problem
The Hierarchical Reasoning Model (HRM) used two small networks recursing at different frequencies, trained on about 1,000 examples, and beat LLMs on Sudoku, Maze and ARC-AGI. The author notes HRM was not well understood and possibly suboptimal.
Method
Tiny Recursive Model (TRM) keeps the core idea, recursion, and drops the rest: a single tiny network with only 2 layers applied repeatedly to refine its reasoning state and answer.
Key results
From the abstract: with 7M parameters, TRM reaches 45% test accuracy on ARC-AGI-1 and 8% on ARC-AGI-2. That’s higher than most LLMs, including DeepSeek R1, o3-mini and Gemini 2.5 Pro, with less than 0.01% of their parameters, and better generalisation than HRM.
Limitations
There’s no dedicated limitations section in the extracted text. TRM is a puzzle-solving network, not a general-purpose language model, so the comparison with LLMs is on these puzzle benchmarks only.
#reasoning#small-models#puzzles
Sources
- Less is More: Recursive Reasoning with Tiny Networks (arXiv:2510.04871) — arXiv
- Hugging Face Daily Papers, 2025-10-08 — Hugging Face
- Less is More: Recursive Reasoning with Tiny Networks (arXiv HTML full text) — arXiv
This is a summary written from the linked public sources. Figures and numbers are as reported there.