Probabilistic Tiny Recursive Model
The article discusses the introduction of a new framework called Probabilistic Tiny Recursive Model (PTRM) designed to enhance the performance of Tiny Recursive Models (TRM) in solving complex reasoning tasks. PTRM incorporates stochastic exploration through Gaussian noise, allowing for improved accuracy without the need for retraining. The framework demonstrates significant accuracy gains across various benchmarks while maintaining a low parameter count.
- ▪Probabilistic Tiny Recursive Model (PTRM) addresses limitations of traditional Tiny Recursive Models (TRM) by introducing stochastic exploration.
- ▪PTRM achieves substantial accuracy improvements on benchmarks like Sudoku-Extreme and Pencil Puzzle Bench.
- ▪The model operates with only 7 million parameters, achieving nearly double the accuracy of leading large language models at a fraction of the cost.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.19943 (cs) [Submitted on 19 May 2026] Title:Probabilistic Tiny Recursive Model Authors:Amin Sghaier, Ali Parviz, Alexia Jolicoeur-Martineau View a PDF of the paper titled Probabilistic Tiny Recursive Model, by Amin Sghaier and 2 other authors View PDF HTML (experimental) Abstract:Tiny Recursive Models (TRM) solve complex reasoning tasks with a fraction of the parameters of modern large language models (LLMs) by iteratively refining a latent state and final answer. While powerful, their deterministic recursion can lead to convergence at suboptimal solutions, without escape mechanism. A common workaround relies on task-specific input perturbations at test time combined with answer aggregation via voting.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
