mlsanityData to 7 Oct 2026

SEAL: Steerable Reasoning Calibration of Large Language Models for Free

arXiv 2504.07986 ↗First posted 7 Apr 2025cs.CL cs.AIcited by 65

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

Abstract (arXiv)

Large Language Models (LLMs), such as OpenAI's o1-series have demonstrated compelling capabilities for complex reasoning tasks via the extended chain-of-thought (CoT) reasoning mechanism. However, recent studies reveal substantial redundancy in the CoT reasoning traces, which not only increases inference latency but also negatively impacts model performance by diverting attention to unnecessary reasoning paths. To address this issue, we investigate the internal reasoning structures of LLMs and categorize them into three primary thought types: execution, reflection, and transition thoughts. Moreover, our analysis reveals that excessive reflection and transition thoughts are strongly correlated with failure cases and these thought categories exhibit clear separation in the latent space. Based on these, we introduce SEAL (Steerable reasoning calibration), a training-free approach that seamlessly calibrates the CoT process, improving accuracy while demonstrating significant efficiency gains. SEAL consists of an offline stage for extracting the reasoning steering vector in the latent space, followed by an on-the-fly calibration of the reasoning trace through representation intervention using the steering vector. Notably, the steering vector exhibits strong transferability across various tasks. Extensive experiments across multiple models (DeepSeek-R1-Distill and QwQ-32B-Preview) and benchmarks (Math500, GSM8K, LiveCodeBench) validate the effectiveness of SEAL, up to a 11% improvement in accuracy while reducing reasoning tokens by 11.8% to 50.4%. Our code is publicly available at https://github.com/VITA-Group/SEAL.

Key facts. Cited by 65 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 25 of its 35 references that are in the map. First posted on arXiv 7 Apr 2025.

Builds on

25 of 35 references.

Closest ancestors · 4

Same direction, newest first

  1. 2025LLM Pretraining with Continuous Concepts12 Feb
  2. 2025s1: Simple test-time scaling31 Jan
  3. 2025Thoughts Are All Over the Place: On the Underthinking of o1-Like LLMs30 Jan
  4. 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs30 Dec

Other references · 21

  1. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
  2. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  3. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  4. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,203
  5. 2023Let's Verify Step by Stepcited by 3,598
  6. 2024Qwen2 Technical Reportcited by 2,076
  7. 2024LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Codecited by 1,906
  8. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 763
  9. 2023WizardMath: Empowering Mathematical Reasoning for Large Language Models via Reinforced Evol-Instructcited by 618
  10. 2023Scaling Relationship on Learning Mathematical Reasoning with Large Language Modelscited by 417
  11. 2024Agentless: Demystifying LLM-based Software Engineering Agentscited by 400
  12. 2023Describe, Explain, Plan and Select: Interactive Planning with Large Language Models Enables Open-World Multi-Task Agentscited by 396
9 more
  1. 2023On the Planning Abilities of Large Language Models : A Critical Investigationcited by 308
  2. 2024Large Language Models for Mathematical Reasoning: Progresses and Challengescited by 283
  3. 2025AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoderscited by 199
  4. 2024Do Large Language Models Latently Perform Multi-Hop Reasoning?cited by 165
  5. 2024NATURAL PLAN: Benchmarking LLMs on Natural Language Planningcited by 125
  6. 2024Debug like a Human: A Large Language Model Debugger via Verifying Runtime Execution Step-by-stepcited by 123
  7. 2024Fantastic Semantics and Where to Find Them: Investigating Which Layers of Generative LLMs Reflect Lexical Semanticscited by 43
  8. 2025Taxonomy, Opportunities, and Challenges of Representation Engineering for Large Language Modelscited by 29
  9. 2024Distributional reasoning in LLMs: Parallel reasoning processes in multi-hop reasoningcited by 21

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 38; 2026: 27.

65 papers.

Sort:

By area:

  1. 2025Dynamic Early Exit in Reasoning Modelscited by 217
  2. 2025Reasoning Beyond Language: A Comprehensive Survey on Latent Chain-of-Thought Reasoningcited by 72
  3. 2025Don't Overthink It: A Survey of Efficient R1-style Large Reasoning Modelscited by 36
  4. 2025Mitigating Overthinking in Large Reasoning Models via Manifold Steeringcited by 32
  5. 2025Steering LLM Thinking with Budget Guidancecited by 30
  6. 2025R1-Compress: Long Chain-of-Thought Compression via Chunk Compression and Searchcited by 29
  7. 2025Beyond Prompt Engineering: Robust Behavior Control in LLMs via Steering Target Atomscited by 23
  8. 2025Activation Steering for Chain-of-Thought Compressioncited by 22
  9. 2025ThinKV: Thought-Adaptive KV Cache Compression for Efficient Reasoning Modelscited by 21
  10. 2025The Geometry of Reasoning: Flowing Logics in Representation Spacecited by 21
  11. 2025Early Stopping Chain-of-thoughts in Large Language Modelscited by 19
  12. 2026Does Your Reasoning Model Implicitly Know When to Stop Thinking?cited by 17
38 more
  1. 2025Two Experts Are All You Need for Steering Thinking: Reinforcing Cognitive Effort in MoE Reasoning Models Without Additional Trainingcited by 16
  2. 2025Can Aha Moments Be Fake? Towards Quantifying Decorative and True Thinking in Chain-of-Thoughtcited by 14
  3. 2025Concise Reasoning, Big Gains: Pruning Long Reasoning Trace with Difficulty-Aware Promptingcited by 13
  4. 2025On Reasoning Strength Planning in Large Reasoning Modelscited by 13
  5. 2026WS-GRPO: Weakly-Supervised Group-Relative Policy Optimization for Rollout-Efficient Reasoningcited by 12
  6. 2026PERSONA: Dynamic and Compositional Inference-Time Personality Control via Activation Vector Algebracited by 11
  7. 2025The First Impression Problem: Internal Bias Triggers Overthinking in Reasoning Modelscited by 10
  8. 2025Efficient Reasoning Through Suppression of Self-Affirmation Reflections in Large Reasoning Modelscited by 10
  9. 2025Understanding and Steering the Cognitive Behaviors of Reasoning Models at Test-Timecited by 10
  10. 2025Exploring and Exploiting the Inherent Efficiency within Large Reasoning Models for Self-Guided Efficiency Enhancementcited by 8
  11. 2025EasyEdit2: An Easy-to-use Steering Framework for Editing Large Language Modelscited by 7
  12. 2025Where Did This Sentence Come From? Tracing Provenance in LLM Reasoning Distillationcited by 6
  13. 2026ATLAS: Verifier-Guided Adaptive Latent Activation Steering for Efficient LLM Reasoningcited by 6
  14. 2025Thinking-Free Policy Initialization Makes Distilled Reasoning Models More Effective and Efficient Reasonerscited by 5
  15. 2025SkipKV: Selective Skipping of KV Generation and Storage for Efficient Inference with Large Reasoning Modelscited by 5
  16. 2025Fantastic Reasoning Behaviors and Where to Find Them: Unsupervised Discovery of the Reasoning Processcited by 5
  17. 2026Towards a Mechanistic Understanding of Large Reasoning Models: A Survey of Training, Inference, and Failurescited by 5
  18. 2026The Hidden Signal of Verifier Strictness: Controlling and Improving Step-Wise Verification via Selective Latent Steeringcited by 5
  19. 2025CyclicReflex: Improving Reasoning Models via Cyclical Reflection Token Schedulingcited by 4
  20. 2026Beyond Scalars: Evaluating and Understanding LLM Reasoning via Geometric Progress and Stabilitycited by 4
  21. 2026Efficient Reasoning with Balanced Thinkingcited by 4
  22. 2025OptScale: Probabilistic Optimality for Inference-time Scalingcited by 3
  23. 2025Answering the Unanswerable Is to Err Knowingly: Analyzing and Mitigating Abstention Failures in Large Reasoning Modelscited by 3
  24. 2025LAWCAT: Efficient Distillation from Quadratic to Linear Attention with Convolution across Tokens for Long Context Modelingcited by 3
  25. 2025Modeling Hierarchical Thinking in Large Reasoning Modelscited by 3
  26. 2025ReflCtrl: Controlling LLM Reflection Efficiently via Representation Engineeringcited by 3
  27. 2026Internalizing LLM Reasoning via Discovery and Replay of Latent Actionscited by 3
  28. 2026Steering Large Reasoning Models towards Concise Reasoning via Flow Matchingcited by 3
  29. 2026From P(y|x) to P(y): Investigating Reinforcement Learning in Pre-train Spacecited by 3
  30. 2025Mind the Gap: Bridging Thought Leap for Improved Chain-of-Thought Tuningcited by 2
  31. 2025Unveiling the Latent Directions of Reflection in Large Language Modelscited by 2
  32. 2025Distractor Injection Attacks on Large Reasoning Models: Characterization and Defensecited by 2
  33. 2026Composer Vector: Style-steering Symbolic Music Generation in a Latent Spacecited by 2
  34. 2026Think Multilingual, Not Harder: A Data-Efficient Framework for Teaching Reasoning Models to Code-Switchcited by 2
  35. 2026The Shape of Overthinking: Backtracking Bursts in Long Reasoning Tracescited by 2
  36. 2026DLWM: Diverse Latent World Models for Efficient Multimodal Reasoningcited by 2
  37. 2025Pedagogy-R1: Pedagogically-Aligned Reasoning Model with Balanced Educational Benchmarkcited by 1
  38. 2025RepV: Safety-Separable Latent Spaces for Scalable Neurosymbolic Plan Verificationcited by 1

50 most cited of 65 All 65 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2025Beyond Chains of Thought: Benchmarking Latent-Space Reasoning Abilities in Large Language Models
  2. 2025Thought calibration: Efficient and confident test-time scaling
  3. 2023Chain-of-Thought Hub: A Continuous Effort to Measure Large Language Models' Reasoning Performance
  4. 2025Understanding Reasoning in Thinking Language Models via Steering Vectors
  5. 2025Reasoning Models Better Express Their Confidence
  6. 2025Think Consistently, Reason Efficiently: Energy-Based Calibration for Implicit Chain-of-Thought
  7. 2025Deep Hidden Cognition Facilitates Reliable Chain-of-Thought Reasoning
  8. 2026On the Generalization of Steering Vectors for Chain-of-Thought Faithfulness