# Test-time Adaptation of Tiny Recursive Models

arXiv: https://arxiv.org/abs/2511.02886 · HTML: https://mlsanity.com/paper/2511.02886/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Builds on 2 of its 17 references that are in the map.
- First posted on arXiv 4 Nov 2025.

## Builds on (most cited first, 2 of 2)

- [Hierarchical Reasoning Model](https://mlsanity.com/paper/2506.21734/) (2025)
- [Searching Latent Program Spaces](https://mlsanity.com/paper/2411.08706/) (2024)

## Built on by (most cited first, 5 of 5)

- [Tiny Recursive Models on ARC-AGI-1: Inductive Biases, Identity Conditioning, and Test-Time Compute](https://mlsanity.com/paper/2512.11847/) (2025)
- [Loop the Loopies!](https://mlsanity.com/paper/2607.16051/) (2026)
- [ARC Prize 2025: Technical Report](https://mlsanity.com/paper/2601.10904/) (2026)
- [Boosting Inference with Guided Reasoning: Stochastic Exploration for Recursive Models](https://mlsanity.com/paper/2605.25230/) (2026)
- [TraceViT: Grounded Trace Supervision for Visual Abstract Reasoning](https://mlsanity.com/paper/2607.29586/) (2026)

## Abstract (arXiv)

Prior to the close of the 2025 ARC Prize competition, the leading open source approach - known as TRM, or Tiny Recursive Models - involved training a 7M parameter recursive neural network on augmented variants of ARC tasks. That approach scored approximately 7.8% on the public ARC AGI II evaluation set, but required a level of compute far in excess of what is allowed during the competition. This paper shows that, by starting from a tiny recursive model that has been pre-trained on public ARC tasks, one can efficiently fine-tune on competition tasks within the allowed compute limits. Specifically, a model was pre-trained on 1,280 public tasks for 700k+ optimizer steps over 48 hours on 4xH100 SXM GPUs to obtain a ~10% score on the public evaluation set. That model was then post-trained in just 12,500 gradient steps during the competition to reach a score of 6.67% on semi-private evaluation tasks. Notably, such post-training performance is achieved by full-fine tuning of the tiny model, not LoRA fine-tuning or fine-tuning of task embeddings alone.