mlsanityData to 7 Oct 2026

Dynamic Early Exit in Reasoning Models

arXiv 2504.15895 ↗First posted 22 Apr 2025cs.CL cs.AIcited by 217

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

also in 1 more
Abstract (arXiv)

Recent advances in large reasoning language models (LRLMs) rely on test-time scaling, which extends long chain-of-thought (CoT) generation to solve complex tasks. However, overthinking in long CoT not only slows down the efficiency of problem solving, but also risks accuracy loss due to the extremely detailed or redundant reasoning steps. We propose a simple yet effective method that allows LLMs to self-truncate CoT sequences by early exit during generation. Instead of relying on fixed heuristics, the proposed method monitors model behavior at potential reasoning transition points and dynamically terminates the next reasoning chain's generation when the model exhibits high confidence in a trial answer. Our method requires no additional training and can be seamlessly integrated into existing o1-like reasoning LLMs. Experiments on 10 reasoning benchmarks (e.g., GSM8K, MATH-500, AMC, GPQA, AIME and LiveCodeBench) show that the proposed method is consistently effective on 11 cutting-edge reasoning LLMs of varying series and sizes, reducing the length of CoT sequences by an average of 19.1% to 80.1% while improving accuracy by 0.3% to 5.0%.

Key facts. Cited by 217 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 66 of its 98 references that are in the map. First posted on arXiv 22 Apr 2025.

Builds on

66 of 98 references.

Closest ancestors · 29

Same direction, newest first

  1. 2025Exploring the System 1 Thinking Capability of Large Reasoning Models14 Apr
  2. 2025Reasoning Models Can Be Effective Without Thinking14 Apr
  3. 2025SEAL: Steerable Reasoning Calibration of Large Language Models for Free7 Apr
  4. 2025Reasoning Models Know When They're Right: Probing Hidden States for Self-Verification7 Apr
  5. 2025Harnessing the Reasoning Economy: A Survey of Efficient Reasoning for Large Language Models31 Mar
  6. 2025Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models20 Mar
  7. 2025L1: Controlling How Long A Reasoning Model Thinks With Reinforcement Learning6 Mar
  8. 2025DAST: Difficulty-Adaptive Slow-Thinking for Large Reasoning Models6 Mar
  9. 2025How Well do LLMs Compress Their Own Chain-of-Thought? A Token Complexity Approach3 Mar
  10. 2025Self-Training Elicits Concise Reasoning in Large Language Models27 Feb
  11. 2025Reasoning with Latent Thoughts: On the Power of Looped Transformers24 Feb
  12. 2025TokenSkip: Controllable Chain-of-Thought Compression in LLMs17 Feb
17 more
  1. 2025CoT-Valve: Length-Compressible Chain-of-Thought Tuning13 Feb
  2. 2025The Danger of Overthinking: Examining the Reasoning-Action Dilemma in Agentic Tasks12 Feb
  3. 2025Demystifying Long Chain-of-Thought Reasoning in LLMs5 Feb
  4. 2025Token Assorted: Mixing Latent and Text Tokens for Improved Language Model Reasoning5 Feb
  5. 2025s1: Simple test-time scaling31 Jan
  6. 2025Reward-Guided Speculative Decoding for Efficient LLM Reasoning31 Jan
  7. 2025Efficient Reasoning with Hidden Thinking31 Jan
  8. 2025O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruning22 Jan
  9. 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs30 Dec
  10. 2024Token-Budget-Aware LLM Reasoning24 Dec
  11. 2024Compressed Chain of Thought: Efficient Reasoning Through Dense Representations17 Dec
  12. 2024C3oT: Generating Shorter Chain-of-Thought without Compromising Effectiveness16 Dec
  13. 2024Adaptive Inference-Time Compute: LLMs Can Predict if They Can Do Better, Even Mid-Generation3 Oct
  14. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters6 Aug
  15. 2024Escape Sky-high Cost: Early-stopping Self-Consistency for Multi-step Reasoning19 Jan
  16. 2023Let's Sample Step by Step: Adaptive-Consistency for Efficient Reasoning and Coding with LLMs19 May
  17. 2023Self-Evaluation Guided Beam Search for Reasoning1 May

Other references · 37

  1. 2023GPT-4 Technical Reportcited by 18,132
  2. 2022Training language models to follow instructions with human feedbackcited by 15,712
  3. 2017Proximal Policy Optimization Algorithmscited by 14,441
  4. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
  5. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  6. 2021Evaluating Large Language Models Trained on Codecited by 8,053
  7. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,767
  8. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,203
  9. 2021Finetuned Language Models Are Zero-Shot Learnerscited by 3,747
  10. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,634
  11. 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,748
  12. 2019HuggingFace's Transformers: State-of-the-art Natural Language Processingcited by 2,201
25 more
  1. 2024LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Codecited by 1,906
  2. 2025Understanding R1-Zero-Like Training: A Critical Perspectivecited by 1,276
  3. 2025Kimi k1.5: Scaling Reinforcement Learning with LLMscited by 1,193
  4. 2024OlympiadBench: A Challenging Benchmark for Promoting AGI with Olympiad-Level Bilingual Multimodal Scientific Problemscited by 1,099
  5. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 763
  6. 2024RouteLLM: Learning to Route LLMs with Preference Datacited by 633
  7. 2024BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructionscited by 527
  8. 2025LIMO: Less is More for Reasoningcited by 488
  9. 2025Towards Large Reasoning Models: A Survey of Reinforced Reasoning with Large Language Modelscited by 291
  10. 2025Chain of Draft: Thinking Faster by Writing Lesscited by 290
  11. 2025Training Language Models to Reason Efficientlycited by 242
  12. 2025CODI: Compressing Chain-of-Thought into Continuous Space via Self-Distillationcited by 216
  13. 2025When More is Less: Understanding Chain-of-Thought Length in LLMscited by 182
  14. 2024Distilling System 2 into System 1cited by 152
  15. 2025Optimizing Test-Time Compute via Meta Reinforcement Fine-Tuningcited by 121
  16. 2025Sketch-of-Thought: Efficient LLM Reasoning with Adaptive Cognitive-Inspired Sketchingcited by 116
  17. 2025SoftCoT: Soft Chain-of-Thought for Efficient Reasoning with LLMscited by 110
  18. 2024Group Robust Preference Optimization in Reward-free RLHFcited by 94
  19. 2024Learning to Route LLMs with Confidence Tokenscited by 66
  20. 2024The Benefits of a Concise Chain of Thought on Problem-Solving in Large Language Modelscited by 62
  21. 2024Can Language Models Learn to Skip Steps?cited by 59
  22. 2024Unlocking the Capabilities of Thought: A Reasoning Boundary Framework to Quantify and Optimize Chain-of-Thoughtcited by 56
  23. 2025Stepwise Perplexity-Guided Refinement for Efficient Chain-of-Thought Reasoning in Large Language Modelscited by 51
  24. 2025Deconstructing Long Chain-of-Thought: A Structured Reasoning Optimization Framework for Long CoT Distillationcited by 39
  25. 2025Confident or Seek Stronger: Exploring Uncertainty-Based On-device LLM Routing From Benchmarking to Generalizationcited by 29

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 104; 2026: 113.

217 papers.

Sort:

By area:

  1. 2025Deep Think with Confidencecited by 199
  2. 2025AbstentionBench: Reasoning LLMs Fail on Unanswerable Questionscited by 129
  3. 2025Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computationcited by 120
  4. 2026Self-Distilled RLVRcited by 109
  5. 2025AdaCoT: Pareto-Optimal Adaptive Chain-of-Thought Triggering via Reinforcement Learningcited by 86
  6. 2025Don't Overthink it. Preferring Shorter Thinking Chains for Improved LLM Reasoningcited by 82
  7. 2025Think Silently, Think Fast: Dynamic Latent Compression of LLM Reasoning Chainscited by 81
  8. 2025Sample More to Think Less: Group Filtered Policy Optimization for Concise Reasoningcited by 75
  9. 2025Wait, We Don't Need to "Wait"! Removing Thinking Tokens Improves Reasoning Efficiencycited by 65
  10. 2025Learning When to Think: Shaping Adaptive Reasoning in R1-Style Models via Multi-Stage RLcited by 59
  11. 2025SelfBudgeter: Adaptive Token Allocation for Efficient LLM Reasoningcited by 47
  12. 2025Answer Convergence as a Signal for Early Stopping in Reasoningcited by 38
38 more
  1. 2025Optimizing Length Compression in Large Reasoning Modelscited by 37
  2. 2025Don't Overthink It: A Survey of Efficient R1-style Large Reasoning Modelscited by 36
  3. 2025ConCISE: Confidence-guided Compression in Step-by-step Efficient Reasoningcited by 33
  4. 2025Think or Not? Exploring Thinking Efficiency in Large Reasoning Models via an Information-Theoretic Lenscited by 32
  5. 2025Mitigating Overthinking in Large Reasoning Models via Manifold Steeringcited by 32
  6. 2025Steering LLM Thinking with Budget Guidancecited by 30
  7. 2025DRP: Distilled Reasoning Pruning with Skill-aware Step Decomposition for Efficient Large Reasoning Modelscited by 29
  8. 2025R1-Compress: Long Chain-of-Thought Compression via Chunk Compression and Searchcited by 29
  9. 2025Optimizing Anytime Reasoning via Budget Relative Policy Optimizationcited by 27
  10. 2025Let LRMs Break Free from Overthinking via Self-Braking Tuningcited by 27
  11. 2025Stable Reinforcement Learning for Efficient Reasoningcited by 27
  12. 2025AlphaOne: Reasoning Models Thinking Slow and Fast at Test Timecited by 27
  13. 2025Promoting Efficient Reasoning with Verifiable Stepwise Rewardcited by 27
  14. 2025When to Continue Thinking: Adaptive Thinking Mode Switching for Efficient Reasoningcited by 26
  15. 2025Shorten After You're Right: Lazy Length Penalties for Reasoning RLcited by 25
  16. 2025VeriThinker: Learning to Verify Makes Reasoning Model Efficientcited by 24
  17. 2025Entropy After </Think> for reasoning model early exitingcited by 24
  18. 2025Activation Steering for Chain-of-Thought Compressioncited by 22
  19. 2025Thought calibration: Efficient and confident test-time scalingcited by 20
  20. 2025THINK-Bench: Evaluating Thinking Efficiency and Chain-of-Thought Quality of Large Reasoning Modelscited by 18
  21. 2025Towards Concise and Adaptive Thinking in Large Reasoning Models: A Surveycited by 18
  22. 2025The Evolution of Thought: Tracking LLM Overthinking via Reasoning Dynamics Analysiscited by 18
  23. 2025The Price of a Second Thought: On the Evaluation of Reasoning Efficiency in Large Language Modelscited by 17
  24. 2025OThink-R1: Intrinsic Fast/Slow Thinking Mode Switching for Over-Reasoning Mitigationcited by 17
  25. 2025MUR: Momentum Uncertainty guided Reasoning for Large Language Modelscited by 17
  26. 2026Does Your Reasoning Model Implicitly Know When to Stop Thinking?cited by 17
  27. 2025AgentTTS: Large Language Model Agent for Test-time Compute-optimal Scaling Strategy in Complex Taskscited by 16
  28. 2025DRPO: Efficient Reasoning via Decoupled Reward Policy Optimizationcited by 16
  29. 2025ARES: Multimodal Adaptive Reasoning via Difficulty-Aware Token-Level Entropy Shapingcited by 16
  30. 2025Lost at the Beginning of Reasoningcited by 15
  31. 2025Stop When Enough: Adaptive Early-Stopping for Chain-of-Thought Reasoningcited by 15
  32. 2025TrimR: Verifier-based Training-Free Thinking Compression for Efficient Test-Time Scalingcited by 14
  33. 2025Can Aha Moments Be Fake? Towards Quantifying Decorative and True Thinking in Chain-of-Thoughtcited by 14
  34. 2025Aware First, Think Less: Dynamic Boundary Self-Awareness Drives Extreme Reasoning Efficiency in Large Language Modelscited by 13
  35. 2025Leash: Adaptive Length Penalty and Reward Shaping for Efficient Large Reasoning Modelcited by 13
  36. 2026Parallel-Probe: Towards Efficient Parallel Thinking via 2D Probingcited by 13
  37. 2025Do Thinking Tokens Help or Trap? Towards More Efficient Large Reasoning Modelcited by 12
  38. 2025Test-time Prompt Interventioncited by 12

50 most cited of 217 All 217 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2025Early Stopping Chain-of-thoughts in Large Language Models
  2. 2025Towards Thinking-Optimal Scaling of Test-Time Compute for LLM Reasoning
  3. 2026</think> Doesn't Stop Reasoning: Analysis of Spurious CoT Termination
  4. 2025One Missing Piece for Open-Source Reasoning Models: A Dataset to Mitigate Cold-Starting Short CoT LLMs in RL
  5. 2025Why Does Reasoning Length Converge? Unveiling the Underfitting-Overfitting Trade-off in Chain-of-Thought
  6. 2025Efficient Reasoning Models: A Survey
  7. 2025Efficient Long CoT Reasoning in Small Language Models
  8. 2025Efficient Inference for Large Reasoning Models: A Survey