mlsanityData to 7 Oct 2026

The Price of a Second Thought: On the Evaluation of Reasoning Efficiency in Large Language Models

arXiv 2505.22017 ↗First posted 28 May 2025cs.CLcited by 17

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

Abstract (arXiv)

Recent thinking models trained with reinforcement learning and backward-checking CoT often suffer from overthinking: they produce excessively long outputs even on simple problems, wasting computation. Existing evaluations, based on token efficiency, give an incomplete view as they neglect problem difficulty and intermediate computation costs. We formalize reasoning efficiency as a relative measure between thinking and instruct models, treating instruct models as the minimal-effort baseline. A systematic study across four thinking models and multiple benchmarks reveals two consistent patterns: (i) instruct models achieve higher efficiency overall, and (ii) problem difficulty affects efficiency, with thinking models wasting computation on easy problems but providing value on harder ones. Building on this insight, we propose COTHINK, a simple two-stage pipeline: an instruct model drafts a brief outline, and a thinking model expands it. On GSM8K, MATH500, and AIME24, COTHINK cuts token usage by 21.1% while keeping accuracy on four thinking models, and remains competitive with strong efficiency baselines.

Key facts. Cited by 17 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 41 of its 49 references that are in the map. First posted on arXiv 28 May 2025.

Builds on

41 of 49 references.

Closest ancestors · 24

Same direction, newest first

  1. 2025Think Only When You Need with Large Hybrid-Reasoning Models20 May
  2. 2025FlashThink: An Early Exit Method For Efficient Reasoning20 May
  3. 2025AdaptThink: Reasoning Models Can Learn When to Think19 May
  4. 2025Lost in Transmission: When and Why LLMs Fail to Reason Globally13 May
  5. 2025Efficient Reasoning for LLMs through Speculative Chain-of-Thought27 Apr
  6. 2025Dynamic Early Exit in Reasoning Models22 Apr
  7. 2025Thought Manipulation: External Thought Can Be Efficient for Large Reasoning Models18 Apr
  8. 2025Efficient Reasoning Models: A Survey15 Apr
  9. 2025Reasoning Models Can Be Effective Without Thinking14 Apr
  10. 2025Reasoning Models Know When They're Right: Probing Hidden States for Self-Verification7 Apr
  11. 2025When To Solve, When To Verify: Compute-Optimal Problem Solving and Generative Verification for LLM Reasoning1 Apr
  12. 2025A Survey of Efficient Reasoning for Large Reasoning Models: Language, Multimodality, and Beyond27 Mar
12 more
  1. 2025Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models20 Mar
  2. 2025InftyThink: Breaking the Length Limits of Long-Context Reasoning in Large Language Models9 Mar
  3. 2025L1: Controlling How Long A Reasoning Model Thinks With Reinforcement Learning6 Mar
  4. 2025Towards Thinking-Optimal Scaling of Test-Time Compute for LLM Reasoning25 Feb
  5. 2025TokenSkip: Controllable Chain-of-Thought Compression in LLMs17 Feb
  6. 2025Thoughts Are All Over the Place: On the Underthinking of o1-Like LLMs30 Jan
  7. 2025O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruning22 Jan
  8. 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs30 Dec
  9. 2024Token-Budget-Aware LLM Reasoning24 Dec
  10. 2024C3oT: Generating Shorter Chain-of-Thought without Compromising Effectiveness16 Dec
  11. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters6 Aug
  12. 2023Skeleton-of-Thought: Prompting LLMs for Efficient Parallel Generation28 Jul

Other references · 17

  1. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  2. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  3. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,203
  4. 2024Qwen2.5 Technical Reportcited by 4,713
  5. 2025DAPO: An Open-Source LLM Reinforcement Learning System at Scalecited by 2,678
  6. 2023Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Promptingcited by 1,081
  7. 2025Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?cited by 948
  8. 2022Decomposed Prompting: A Modular Approach for Solving Complex Taskscited by 554
  9. 2020Efficient Content-Based Sparse Attention with Routing Transformerscited by 469
  10. 2025Reasoning Models Don't Always Say What They Thinkcited by 388
  11. 2025Chain of Draft: Thinking Faster by Writing Lesscited by 290
  12. 2025Training Language Models to Reason Efficientlycited by 242
5 more
  1. 2021Inductive Biases and Variable Creation in Self-Attention Mechanismscited by 144
  2. 2025Optimizing Test-Time Compute via Meta Reinforcement Fine-Tuningcited by 121
  3. 2025Sketch-of-Thought: Efficient LLM Reasoning with Adaptive Cognitive-Inspired Sketchingcited by 116
  4. 2025Toward Generalizable Evaluation in the LLM Era: A Survey Beyond Benchmarkscited by 26
  5. 2023Prompt Sketching for Large Language Modelscited by 3

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 7; 2026: 10.

17 papers.

Sort:

By area:

  1. 2025Don't Overthink It: A Survey of Efficient R1-style Large Reasoning Modelscited by 36
  2. 2025Towards Concise and Adaptive Thinking in Large Reasoning Models: A Surveycited by 18
  3. 2025SABER: Switchable and Balanced Training for Efficient LLM Reasoningcited by 12
  4. 2025The Markovian Thinker: Architecture-Agnostic Linear Scaling of Reasoningcited by 12
  5. 2025From Efficiency to Adaptivity: A Deeper Look at Adaptive Reasoning in Large Language Modelscited by 11
  6. 2026Neural Chain-of-Thought Search: Searching the Optimal Reasoning Path to Enhance Large Language Modelscited by 7
  7. 2025SLR: Automated Synthesis for Scalable Logical Reasoningcited by 6
  8. 2026Probing the Trajectories of Reasoning Traces in Large Language Modelscited by 6
  9. 2026Accordion-Thinking: Self-Regulated Step Summaries for Efficient and Readable LLM Reasoningcited by 4
  10. 2026System 1&2 Synergy via Dynamic Model Interpolationcited by 3
  11. 2025Expanding Computation Spaces of LLMs at Inference Timecited by 2
  12. 2026PACE: Prefix-Protected and Difficulty-Aware Compression for Efficient Reasoningcited by 2
5 more
  1. 2026Stop Unnecessary Reflection: Training LRMs for Efficient Reasoning with Adaptive Reflection and Length Coordinated Penaltycited by 2
  2. 2026TRiMS: Real-Time Tracking of Minimal Sufficient Length for Efficient Reasoning via RLcited by 2
  3. 2026Reasoning about Reasoning: BAPO Bounds on Chain-of-Thought Token Complexity in LLMscited by 1
  4. 2026When Less is Enough: Efficient Inference via Collaborative Reasoningcited by 0
  5. 2026Revisiting Complete Reasoning Traces for Post-Trainingcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026The Art of Efficient Reasoning: Data, Reward, and Optimization
  2. 2025Exploring and Exploiting the Inherent Efficiency within Large Reasoning Models for Self-Guided Efficiency Enhancement
  3. 2026When More Thinking Hurts: Overthinking in LLM Test-Time Compute Scaling
  4. 2025Large Reasoning Models are not thinking straight: on the unreliability of thinking trajectories
  5. 2026Batched Contextual Reinforcement: A Task-Scaling Law for Efficient Reasoning
  6. 2025How Far Are We from Optimal Reasoning Efficiency?
  7. 2025Performative Thinking? The Brittle Correlation Between CoT Length and Problem Complexity
  8. 2025Not All Thoughts are Generated Equal: Efficient LLM Reasoning via Multi-Turn Reinforcement Learning