When to Vote, When to Rewrite: Disagreement-Guided Strategy Routing for Test-Time Scaling
On the map: LLM reasoning → Test-time compute and efficient thinking
Abstract (arXiv)
Large Reasoning Models (LRMs) achieve strong performance on mathematical reasoning tasks but remain unreliable on challenging instances. Existing test-time scaling methods, such as repeated sampling, self-correction, and tree search, improve performance at the cost of increased computation, yet often exhibit diminishing returns on hard problems. We observe that output disagreement is strongly correlated with instance difficulty and prediction correctness, providing a useful signal for guiding instance-level strategy selection at test time. Based on this insight, we propose a training-free framework that formulates test-time scaling as an instance-level routing problem, rather than allocating more computation within a single strategy, dynamically selecting among different scaling strategies based on output disagreement. The framework applies lightweight resolution for consistent cases, majority voting for moderate disagreement, and rewriting-based reformulation for highly ambiguous instances. Experiments on seven mathematical benchmarks and three models show that our method improves accuracy by 3% - 7% while reducing sampling cost compared to existing approaches.
Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 38 of its 42 references that are in the map. First posted on arXiv 29 Apr 2026.
Builds on
38 of 42 references.
Closest ancestors · 14
Same direction, newest first
- 2026When Is Thinking Enough? Early Exit via Sufficiency Assessment for Efficient Reasoning8 Apr
- 2025The Art of Scaling Test-Time Compute for Large Language Models1 Dec
- 2025Test-Time Scaling in Reasoning Models Is Not Effective for Knowledge-Intensive Tasks Yet8 Sep
- 2025Don't Overthink it. Preferring Shorter Thinking Chains for Improved LLM Reasoning23 May
- 2025A Survey on Test-Time Scaling in Large Language Models: What, How, Where, and How Well?31 Mar
- 2025Linguistic Generalizability of Test-Time Scaling in Mathematical Reasoning24 Feb
- 2025A Survey of Test-Time Compute: From Intuitive Inference to Deliberate Reasoning5 Jan
- 2024Forest-of-Thought: Scaling Test-Time Compute for Enhancing LLM Reasoning12 Dec
- 2024Embedding Self-Correction as an Inherent Ability in Large Language Models for Enhanced Mathematical Reasoning14 Oct
- 2024Interpretable Contrastive Monte Carlo Tree Search Reasoning2 Oct
- 2024Training Language Models to Self-Correct via Reinforcement Learning19 Sep
- 2023Just Ask One More Time! Self-Agreement Improves Reasoning of Language Models in (Almost) All Scenarios14 Nov
2 more
Other references · 24
- 2021Training Verifiers to Solve Math Word Problemscited by 9,743
- 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
- 2021Evaluating Large Language Models Trained on Codecited by 8,053
- 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,601
- 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,203
- 2023Reflexion: Language Agents with Verbal Reinforcement Learningcited by 4,064
- 2021Program Synthesis with Large Language Modelscited by 3,509
- 2024OlympiadBench: A Challenging Benchmark for Promoting AGI with Olympiad-Level Bilingual Multimodal Scientific Problemscited by 1,099
- 2024Qwen2.5-Math Technical Report: Toward Mathematical Expert Model via Self-Improvementcited by 1,057
- 2023Teaching Large Language Models to Self-Debugcited by 963
- 2025Skywork-Reward-V2: Scaling Preference Data Curation via Human-AI Synergycited by 204
- 2023Evaluating the Performance of Large Language Models on GAOKAO Benchmarkcited by 182
12 more
- 2024More Agents Is All You Needcited by 174
- 2024Accessing GPT-4 level Mathematical Olympiad Solutions via Monte Carlo Tree Self-refine with LLaMa-3 8Bcited by 156
- 2024BoNBoN Alignment for Large Language Models and the Sweetness of Best-of-n Samplingcited by 120
- 2023RewriteLM: An Instruction-Tuned Large Language Model for Text Rewritingcited by 67
- 2024Confidence Matters: Revisiting Intrinsic Self-Correction Capabilities of Large Language Modelscited by 47
- 2024PRewrite: Prompt Rewriting with Reinforcement Learningcited by 43
- 2024LLM-R2: A Large Language Model Enhanced Rule-based Rewrite System for Boosting Query Efficiencycited by 16
- 2024Paraphrase and Solve: Exploring and Exploiting the Impact of Surface Form on Mathematical Reasoning in Large Language Modelscited by 14
- 2025Efficient Prediction of Pass@k Scaling in Large Language Modelscited by 14
- 2025Multidimensional Consistency Improves Reasoning in Language Modelscited by 6
- 2024Putting People in LLMs' Shoes: Generating Better Answers via Question Rewritercited by 3
- 2025RM-PoT: Reformulating Mathematical Problems and Solving via Program of Thoughtscited by 1
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2025Scaling over Scaling: Exploring Test-Time Scaling Plateau in Large Reasoning Models
- 2025ARISE: An Adaptive Resolution-Aware Metric for Test-Time Scaling Evaluation in Large Reasoning Models
- 2026Test-Time Scaling in Reasoning LLMs: Inference Regimes, Evaluation, and Reproducibility
- 2025Mitigating Strategy-Selection Bias in Reasoning for More Effective Test-Time Scaling
- 2025SCALE: Selective Resource Allocation for Overcoming Performance Bottlenecks in Mathematical Test-time Scaling
- 2026OpenDeepThink: Parallel Reasoning via Bradley-Terry Aggregation
- 2026Divergent-Convergent Reasoning: Scaling Test-Time Compute through Structured Solution Synthesis
- 2026Planned Test-Time Scaling with Coordinated Reasoning Paths