mlsanityData to 7 Oct 2026

Efficient Reasoning with Balanced Thinking

arXiv 2603.12372 ↗First posted 12 Mar 2026cs.AI cs.CL cs.LGcited by 4

On the map: LLM reasoning → Test-time compute and efficient thinking

Abstract (arXiv)

Large Reasoning Models (LRMs) have shown remarkable reasoning capabilities, yet they often suffer from overthinking, expending redundant computational steps on simple problems, or underthinking, failing to explore sufficient reasoning paths despite inherent capabilities. These issues lead to inefficiencies and potential inaccuracies, limiting practical deployment in resource-constrained settings. Existing methods to mitigate overthinking, such as suppressing reflective keywords or adjusting reasoning length, may inadvertently induce underthinking, compromising accuracy. Therefore, we propose ReBalance, a training-free framework that achieves efficient reasoning with balanced thinking. ReBalance leverages confidence as a continuous indicator of reasoning dynamics, identifying overthinking through high confidence variance and underthinking via consistent overconfidence. By aggregating hidden states from a small-scale dataset into reasoning mode prototypes, we compute a steering vector to guide LRMs' reasoning trajectories. A dynamic control function modulates this vector's strength and direction based on real-time confidence, pruning redundancy during overthinking, and promoting exploration during underthinking. Extensive experiments conducted on four models ranging from 0.5B to 32B, and across nine benchmarks in math reasoning, general question answering, and coding tasks demonstrate that ReBalance effectively reduces output redundancy while improving accuracy, offering a general, training-free, and plug-and-play strategy for efficient and robust LRM deployment. Project page and code are available at https://rebalance-ai.github.io .

Key facts. Cited by 4 papers in the map (data through 7 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 70 of its 122 references that are in the map. First posted on arXiv 12 Mar 2026.

Builds on

70 of 122 references.

Closest ancestors · 33

Same direction, newest first

  1. 2025Don't Overthink It: A Survey of Efficient R1-style Large Reasoning Models4 Aug
  2. 2025Controlling Thinking Speed in Reasoning Models4 Jul
  3. 2025On Reasoning Strength Planning in Large Reasoning Models10 Jun
  4. 2025Wait, We Don't Need to "Wait"! Removing Thinking Tokens Improves Reasoning Efficiency10 Jun
  5. 2025OThink-R1: Intrinsic Fast/Slow Thinking Mode Switching for Over-Reasoning Mitigation3 Jun
  6. 2025Mitigating Overthinking in Large Reasoning Models via Manifold Steering28 May
  7. 2025AdaCtrl: Towards Adaptive and Controllable Reasoning via Difficulty-Aware Budgeting24 May
  8. 2025Thinking Fast and Right: Balancing Accuracy and Reasoning Length with Adaptive Rewards23 May
  9. 2025Not All Tokens Are What You Need In Thinking23 May
  10. 2025TrimR: Verifier-based Training-Free Thinking Compression for Efficient Test-Time Scaling22 May
  11. 2025FlashThink: An Early Exit Method For Efficient Reasoning20 May
  12. 2025Thinkless: LLM Learns When to Think19 May
21 more
  1. 2025AdaCoT: Pareto-Optimal Adaptive Chain-of-Thought Triggering via Reinforcement Learning17 May
  2. 2025Between Underthinking and Overthinking: An Empirical Study of Reasoning Length and correctness in LLMs30 Apr
  3. 2025Dynamic Early Exit in Reasoning Models22 Apr
  4. 2025Reasoning Models Can Be Effective Without Thinking14 Apr
  5. 2025SEAL: Steerable Reasoning Calibration of Large Language Models for Free7 Apr
  6. 2025Quantization Hurts Reasoning? An Empirical Study on Quantized Reasoning Models7 Apr
  7. 2025When Reasoning Meets Compression: Understanding the Effects of LLMs Compression on Large Reasoning Models2 Apr
  8. 2025Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models20 Mar
  9. 2025Entropy-based Exploration Conduction for Multi-step Reasoning20 Mar
  10. 2025DAST: Difficulty-Adaptive Slow-Thinking for Large Reasoning Models6 Mar
  11. 2025Sampling-Efficient Test-Time Scaling: Self-Estimating the Best-of-N Sampling in Early Decoding3 Mar
  12. 2025How Well do LLMs Compress Their Own Chain-of-Thought? A Token Complexity Approach3 Mar
  13. 2025TokenSkip: Controllable Chain-of-Thought Compression in LLMs17 Feb
  14. 2025FastMCTS: A Simple Sampling Strategy for Data Synthesis17 Feb
  15. 2025Token Assorted: Mixing Latent and Text Tokens for Improved Language Model Reasoning5 Feb
  16. 2025s1: Simple test-time scaling31 Jan
  17. 2025Thoughts Are All Over the Place: On the Underthinking of o1-Like LLMs30 Jan
  18. 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs30 Dec
  19. 2024Compressed Chain of Thought: Efficient Reasoning Through Dense Representations17 Dec
  20. 2024Training Language Models to Self-Correct via Reinforcement Learning19 Sep
  21. 2024Concise Thoughts: Impact of Output Length on LLM Reasoning and Cost29 Jul

Other references · 37

  1. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  2. 2025Qwen3 Technical Reportcited by 8,195
  3. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  4. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,601
  5. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,203
  6. 2019HuggingFace's Transformers: State-of-the-art Natural Language Processingcited by 2,201
  7. 2024LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Codecited by 1,906
  8. 2024OpenAI o1 System Cardcited by 1,220
  9. 2024OlympiadBench: A Challenging Benchmark for Promoting AGI with Olympiad-Level Bilingual Multimodal Scientific Problemscited by 1,099
  10. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 763
  11. 2025Layer by Layer: Uncovering Hidden Representations in Language Modelscited by 293
  12. 2025Chain of Draft: Thinking Faster by Writing Lesscited by 290
25 more
  1. 2025Training Language Models to Reason Efficientlycited by 242
  2. 2023In-context Vectors: Making In Context Learning More Effective and Controllable Through Latent Space Steeringcited by 240
  3. 2024The Impact of Reasoning Step Length on Large Language Modelscited by 152
  4. 2024Distilling System 2 into System 1cited by 152
  5. 2025SoftCoT: Soft Chain-of-Thought for Efficient Reasoning with LLMscited by 110
  6. 2025Soft Thinking: Unlocking the Reasoning Potential of LLMs in Continuous Concept Spacecited by 109
  7. 2025Perception, Reason, Think, and Plan: A Survey on Large Multimodal Reasoning Modelscited by 89
  8. 2025More Thinking, Less Seeing? Assessing Amplified Hallucination in Multimodal Reasoning Modelscited by 84
  9. 2025Reasoning with Sampling: Your Base Model is Smarter Than You Thinkcited by 84
  10. 2023LISA++: An Improved Baseline for Reasoning Segmentation with Large Language Modelcited by 70
  11. 2025Pangu Embedded: An Efficient Dual-system LLM Reasoner with Metacognitioncited by 61
  12. 2024Reducing Hallucinations in Vision-Language Models via Latent Space Steeringcited by 56
  13. 2025Memory Forcing: Spatio-Temporal Memory for Consistent Scene Generation on Minecraftcited by 46
  14. 2024EDT: Improving Large Language Models' Generation by Entropy-based Dynamic Temperature Samplingcited by 42
  15. 2025Deconstructing Long Chain-of-Thought: A Structured Reasoning Optimization Framework for Long CoT Distillationcited by 39
  16. 2025Think When You Need: Self-Adaptive Chain-of-Thought Learningcited by 37
  17. 2024Break the Chain: Large Language Models Can be Shortcut Reasonerscited by 32
  18. 2025Detection and Mitigation of Hallucination in Large Reasoning Models: A Mechanistic Perspectivecited by 32
  19. 2025Towards Reasoning Ability of Small Language Modelscited by 29
  20. 2024Scalable Language Model with Generalized Continual Learningcited by 28
  21. 2025Mitigating Object Hallucinations via Sentence-Level Early Interventioncited by 28
  22. 2025Less Is More, but Where? Dynamic Token Compression via LLM-Guided Keyframe Priorcited by 15
  23. 2025Where to show Demos in Your Prompt: A Positional Bias of In-Context Learningcited by 13
  24. 2025Generalized Decoupled Learning for Enhancing Open-Vocabulary Dense Perceptioncited by 9
  25. 2025Inside-Out: Hidden Factual Knowledge in LLMscited by 5

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 4.

4 papers.

  1. 2026Not All Turns Are Equally Hard: Adaptive Thinking Budgets For Efficient Multi-Turn Reasoning in Agentscited by 3
  2. 2026DyCon: Dynamic Reasoning Control via Evolving Difficulty Modelingcited by 0
  3. 2026Improving Test-Time Scaling with Adaptive Looped Transformerscited by 0
  4. 2026Offline Guidance, Online Reasoning: Reusing LLM Feedback for Small Language Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Think How to Think: Mitigating Overthinking with Autonomous Difficulty Cognition in Large Reasoning Models
  2. 2026Thinking Past the Answer: Evaluating Harmful Overthinking in Large Reasoning Models
  3. 2025Let LRMs Break Free from Overthinking via Self-Braking Tuning
  4. 2026When Is Thinking Enough? Early Exit via Sufficiency Assessment for Efficient Reasoning
  5. 2026Mitigating Overthinking in Large Reasoning Language Models via Reasoning Path Deviation Monitoring
  6. 2025Exploring and Exploiting the Inherent Efficiency within Large Reasoning Models for Self-Guided Efficiency Enhancement
  7. 2025Large Reasoning Models are not thinking straight: on the unreliability of thinking trajectories
  8. 2026RFEval: Benchmarking Reasoning Faithfulness under Counterfactual Reasoning Intervention in Large Reasoning Models