mlsanityData to 14 Oct 2026

The Evolution of Thought: Tracking LLM Overthinking via Reasoning Dynamics Analysis

arXiv 2508.17627 ↗First posted 25 Aug 2025cs.CL cs.AIcited by 18

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

Abstract (arXiv)

Test-time scaling via explicit reasoning trajectories significantly boosts large language model (LLM) performance but often triggers overthinking. To explore this, we analyze reasoning through two lenses: Reasoning Length Dynamics, which reveals a compensatory trade-off between thinking and answer content length that eventually leads to thinking redundancy, and Reasoning Semantic Dynamics, which identifies semantic convergence and repetitive oscillations. These dynamics uncover an instance-specific Reasoning Completion Point (RCP), beyond which computation continues without further performance gain. Since the RCP varies across instances, we propose a Reasoning Completion Point Detector (RCPD), an inference-time early-exit method that identifies the RCP by monitoring the rank dynamics of termination tokens (e.g., </think>). Across AIME and GPQA benchmarks using Qwen3 and DeepSeek-R1, RCPD reduces token usage by up to 44% while preserving accuracy, offering a principled approach to efficient test-time scaling.

Key facts. Cited by 18 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 29 of its 40 references that are in the map. First posted on arXiv 25 Aug 2025.

Builds on

29 of 40 references.

Closest ancestors · 17

Same direction, newest first

  1. 2025gpt-oss-120b & gpt-oss-20b Model Card8 Aug
  2. 2025Efficient Reasoning for Large Reasoning Language Models via Certainty-Guided Reflection Suppression7 Aug
  3. 2025Does Thinking More always Help? Mirage of Test-Time Scaling in Reasoning Models4 Jun
  4. 2025Revisiting Overthinking in Long Chain-of-Thought from the Perspective of Self-Doubt29 May
  5. 2025Between Underthinking and Overthinking: An Empirical Study of Reasoning Length and correctness in LLMs30 Apr
  6. 2025Dynamic Early Exit in Reasoning Models22 Apr
  7. 2025THOUGHTTERMINATOR: Benchmarking, Calibrating, and Mitigating Overthinking in Reasoning Models17 Apr
  8. 2025Harnessing the Reasoning Economy: A Survey of Efficient Reasoning for Large Language Models31 Mar
  9. 2025Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models20 Mar
  10. 2025L1: Controlling How Long A Reasoning Model Thinks With Reinforcement Learning6 Mar
  11. 2025How Well do LLMs Compress Their Own Chain-of-Thought? A Token Complexity Approach3 Mar
  12. 2025TokenSkip: Controllable Chain-of-Thought Compression in LLMs17 Feb
5 more
  1. 2025CoT-Valve: Length-Compressible Chain-of-Thought Tuning13 Feb
  2. 2025s1: Simple test-time scaling31 Jan
  3. 2025O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruning22 Jan
  4. 2024Token-Budget-Aware LLM Reasoning24 Dec
  5. 2024Compressed Chain of Thought: Efficient Reasoning Through Dense Representations17 Dec

Other references · 12

  1. 2025Qwen3 Technical Reportcited by 8,820
  2. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  3. 2023Let's Verify Step by Stepcited by 3,789
  4. 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,879
  5. 2025Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Modelscited by 1,261
  6. 2024OpenAI o1 System Cardcited by 1,244
  7. 2025Kimi k1.5: Scaling Reinforcement Learning with LLMscited by 1,210
  8. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 797
  9. 2025CODI: Compressing Chain-of-Thought into Continuous Space via Self-Distillationcited by 225
  10. 2025When More is Less: Understanding Chain-of-Thought Length in LLMscited by 183
  11. 2024Distilling System 2 into System 1cited by 159
  12. 2025S-GRPO: Early Exit via Reinforcement Learning in Reasoning Modelscited by 87

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 2; 2026: 16.

18 papers.

Sort:

  1. 2025When to Reason: Semantic Router for vLLMcited by 12
  2. 2026Conformal Thinking: Risk Control for Reasoning on a Compute Budgetcited by 10
  3. 2026Latent-GRPO: Group Relative Policy Optimization for Latent Reasoningcited by 10
  4. 2026ATTNPO: Attention-Guided Process Supervision for Efficient Reasoningcited by 9
  5. 2026Prioritize the Process, Not Just the Outcome: Rewarding Latent Thought Trajectories Improves Reasoning in Looped Language Modelscited by 8
  6. 2026Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Modelscited by 7
  7. 2026Early Stopping for Large Reasoning Models via Confidence Dynamicscited by 6
  8. 2026Circular Reasoning: Understanding Self-Reinforcing Loops in Large Reasoning Modelscited by 5
  9. 2026ESTAR: Early-Stopping Token-Aware Reasoning For Efficient Inferencecited by 3
  10. 2026SliceGraph: Mapping Process Isomers in Multi-Run Chain-of-Thought Reasoningcited by 2
  11. 2026The Shape of Overthinking: Backtracking Bursts in Long Reasoning Tracescited by 2
  12. 2025Batch Prompting Suppresses Overthinking Reasoning Under Constraint: How Batch Prompting Suppresses Overthinking in Reasoning Modelscited by 1
6 more
  1. 2026STOP: Structured On-Policy Pruning of Long-Form Reasoning in Low-Data Regimescited by 1
  2. 2026SLAT: Segment-Level Adaptive Trimming for Efficient CoT Reasoningcited by 1
  3. 2026Dynamic Rollout Editing for Reducing Overthinking in RL-Trained Reasoning Modelscited by 1
  4. 2026CUSUM-Shaped Inference-Time Monitoring and Targeted Re-Decoding for Quantized Small Language Model Reasoningcited by 1
  5. 2026Metacognitive Behavioral Tuning of Large Language Models for Multi-Hop Question Answeringcited by 0
  6. 2026A First-Principles Theory of Slow Thinking and Active Perceptioncited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026When More Thinking Hurts: Overthinking in LLM Test-Time Compute Scaling
  2. 2026Mitigating Overthinking in Large Reasoning Language Models via Reasoning Path Deviation Monitoring
  3. 2025Why Does Reasoning Length Converge? Unveiling the Underfitting-Overfitting Trade-off in Chain-of-Thought
  4. 2025Thought calibration: Efficient and confident test-time scaling
  5. 2026Thinking Past the Answer: Evaluating Harmful Overthinking in Large Reasoning Models
  6. 2025Demystifying Reasoning Dynamics with Mutual Information: Thinking Tokens are Information Peaks in LLM Reasoning
  7. 2026Probing the Trajectories of Reasoning Traces in Large Language Models
  8. 2025Large Reasoning Models are not thinking straight: on the unreliability of thinking trajectories