# Wait, Wait, Wait... Why Do Reasoning Models Loop?

arXiv: https://arxiv.org/abs/2512.12895 · HTML: https://mlsanity.com/paper/2512.12895/

## Key facts

- Cited by 26 papers in the map (data through 14 Oct 2026).
- Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning.
- Builds on 12 of its 30 references that are in the map.
- First posted on arXiv 15 Dec 2025.

## Placed in

- LLM reasoning / [Test-time compute and efficient thinking](https://mlsanity.com/lr/test-time-compute-and-efficient-thinking/)

## Builds on (most cited first, 12 of 12)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Qwen2 Technical Report](https://mlsanity.com/paper/2407.10671/) (2024)
- [OpenAI o1 System Card](https://mlsanity.com/paper/2412.16720/) (2024)
- [Hierarchical Neural Story Generation](https://mlsanity.com/paper/1805.04833/) (2018)
- [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://mlsanity.com/paper/1909.05858/) (2019)
- [Phi-4 Technical Report](https://mlsanity.com/paper/2412.08905/) (2024)
- [OpenThoughts: Data Recipes for Reasoning Models](https://mlsanity.com/paper/2506.04178/) (2025)
- [The Danger of Overthinking: Examining the Reasoning-Action Dilemma in Agentic Tasks](https://mlsanity.com/paper/2502.08235/) (2025)
- [Phi-4-reasoning Technical Report](https://mlsanity.com/paper/2504.21318/) (2025)
- [LLMs Can Easily Learn to Reason from Demonstrations Structure, not content, is what matters!](https://mlsanity.com/paper/2502.07374/) (2025)
- [Relating Neural Text Degeneration to Exposure Bias](https://mlsanity.com/paper/2109.08705/) (2021)

## Built on by (most cited first, 26 of 26)

- [Endless Terminals: Scaling RL Environments for Terminal Agents](https://mlsanity.com/paper/2601.16443/) (2026)
- [Rethinking Generalization in Reasoning SFT: A Conditional Analysis on Optimization, Data, and Model Capability](https://mlsanity.com/paper/2604.06628/) (2026)
- [Group Distributionally Robust Optimization-Driven Reinforcement Learning for LLM Reasoning](https://mlsanity.com/paper/2601.19280/) (2026)
- [Quantized Reasoning Models Think They Need to Think Longer, but They Do Not](https://mlsanity.com/paper/2606.00206/) (2026)
- [Mechanistic Evidence for Faithfulness Decay in Chain-of-Thought Reasoning](https://mlsanity.com/paper/2602.11201/) (2026)
- [Measuring Progress in Reasoning Toward Mathematical Discovery with Automatic Verification](https://mlsanity.com/paper/2603.15617/) (2026)
- [ShortOPD: Recovering Pruned LLMs with Short-to-Long On-Policy Distillation](https://mlsanity.com/paper/2607.13124/) (2026)
- [Learning from Partial Chain-of-Thought via Truncated-Reasoning Self-Distillation](https://mlsanity.com/paper/2603.13274/) (2026)
- [Argumentation for Explainable and Globally Contestable Decision Support with LLMs](https://mlsanity.com/paper/2603.14643/) (2026)
- [Certain Head, Uncertain Tail: Expert-Sample for Test-Time Scaling in Fine-Grained MoE](https://mlsanity.com/paper/2602.02443/) (2026)
- [Prefix Sliding for efficient test-time scaling](https://mlsanity.com/paper/2608.26070/) (2026)
- [Distillation Traps and Guards: A Calibration Knob for LLM Distillability](https://mlsanity.com/paper/2604.18963/) (2026)
- [Search for Truth from Reasoning: A Dynamic Representation Editing Framework for Steering LLM Trajectories](https://mlsanity.com/paper/2606.28589/) (2026)
- [Know When to Stop: Segment-Level Credit Assignment for Reducing Overthinking](https://mlsanity.com/paper/2607.00482/) (2026)
- [Spatial Reasoning in LLM Game Agents: Impact of Causal Context and Multi-Step Planning](https://mlsanity.com/paper/2607.22732/) (2026)
- [Fork-and-Flush: Escaping Idea Basins in Autoresearch Agents](https://mlsanity.com/paper/2610.07447/) (2026)
- [A Multi-Agent Framework for Interpreting Multivariate Physiological Time Series](https://mlsanity.com/paper/2603.04142/) (2026)
- [Decoding the Critique Mechanism in Large Reasoning Models](https://mlsanity.com/paper/2603.16331/) (2026)
- [Reasoning Structure of Large Language Models](https://mlsanity.com/paper/2606.03883/) (2026)
- [Data-Centric Benchmarking of Exploit Generation in LLMs: Understanding the Impact of Fine-Tuning](https://mlsanity.com/paper/2606.15123/) (2026)
- [Token Budget Saturation and Mechanistic Early Detection of Reasoning Non-Convergence in Chain-of-Thought Models](https://mlsanity.com/paper/2607.21433/) (2026)
- [Thinking effort aligns between humans and reasoning models in abductive reasoning](https://mlsanity.com/paper/2609.01867/) (2026)
- [LLMs as Post-hoc Auditors of Physiological Plausibility in Symbolic Regression: A Clinician-Evaluated Case Study](https://mlsanity.com/paper/2609.11431/) (2026)
- [What Does Post-Training Change in Multilingual Reasoning?](https://mlsanity.com/paper/2609.37104/) (2026)
- [Spend Bytes on Breadth: Precision-Count Trade-offs for Decode-Time KV Compression in Long Chain-of-Thought Reasoning](https://mlsanity.com/paper/2610.05685/) (2026)
- [Progress-Aware Distillation for Mitigating Stagnation in Small Language Model Agents](https://mlsanity.com/paper/2610.19439/) (2026)

## Abstract (arXiv)

Reasoning models (e.g., DeepSeek-R1) generate long chains of thought to solve harder problems, but they often loop, repeating the same text at low temperatures or with greedy decoding. We study why this happens and what role temperature plays. With open reasoning models, we find that looping is common at low temperature. Larger models tend to loop less, and distilled students loop significantly even when their teachers rarely do. This points to mismatches between the training distribution and the learned model, which we refer to as errors in learning, as a key cause. To understand how such errors cause loops, we introduce a synthetic graph reasoning task and demonstrate two mechanisms. First, risk aversion caused by hardness of learning: when the correct progress-making action is hard to learn but an easy cyclic action is available, the model puts relatively more probability on the cyclic action and gets stuck. Second, even when there is no hardness, Transformers show an inductive bias toward temporally correlated errors, so the same few actions keep being chosen and loops appear. Higher temperature reduces looping by promoting exploration, but it does not fix the errors in learning, so generations remain much longer than necessary at high temperature; in this sense, temperature is a stopgap rather than a holistic solution. We end with a discussion of training-time interventions aimed at directly reducing errors in learning.