# Latent Reasoning in TRMs is Secretly a Policy Improvement Operator

Authors: Arip Asadulaev, Rayan Banerjee, Fakhri Karray, Martin Takac.

Interprets recursive state updates as changes to a model's answer distribution. Trains intermediate updates toward progressively less corrupted puzzle solutions, testing whether recursion improves the answer at each step.

First public: 2025-11-21.
ArXiv initial submission: 2025-11-21.

[Original paper](https://arxiv.org/abs/2511.16886) · [PDF](https://arxiv.org/pdf/2511.16886) · [HTML](https://arxiv.org/html/2511.16886v5) · [Reading-list entry](https://latent-implicit-thinking.github.io/papers/2511.16886/)
[Preprint BibTeX](https://latent-implicit-thinking.github.io/papers/2511.16886/citation.bib) · [Preprint CSL-JSON](https://latent-implicit-thinking.github.io/papers/2511.16886/citation.csl.json) · [JSON record](https://latent-implicit-thinking.github.io/papers/2511.16886/record.json)

## Approaches

- [Recurrent reasoning](https://latent-implicit-thinking.github.io/topics/recurrent-reasoning/): Repeat parts of the model or keep intermediate information in memory while working through a problem. Some methods also learn how many passes to run.
- [Analysis and evaluation](https://latent-implicit-thinking.github.io/topics/implicit-reasoning-evaluation/): Measure how well these methods work, investigate what their internal computation does, and identify where they fail. This includes experiments, studies of model internals, and mathematical results.

Reading-list version 0.12.0; literature checked through 2026-11-02.