mlsanityData to 14 Oct 2026

When to Continue Thinking: Adaptive Thinking Mode Switching for Efficient Reasoning

arXiv 2505.15400 ↗First posted 21 May 2025cs.AI cs.CLcited by 26

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

Abstract (arXiv)

Large reasoning models (LRMs) achieve remarkable performance via long reasoning chains, but often incur excessive computational overhead due to redundant reasoning, especially on simple tasks. In this work, we systematically quantify the upper bounds of LRMs under both Long-Thinking and No-Thinking modes, and uncover the phenomenon of "Internal Self-Recovery Mechanism" where models implicitly supplement reasoning during answer generation. Building on this insight, we propose Adaptive Self-Recovery Reasoning (ASRR), a framework that suppresses unnecessary reasoning and enables implicit recovery. By introducing accuracy-aware length reward regulation, ASRR adaptively allocates reasoning effort according to problem difficulty, achieving high efficiency with negligible performance sacrifice. Experiments across multiple benchmarks and models show that, compared with GRPO, ASRR reduces reasoning budget by up to 32.5% (1.5B) and 25.7% (7B) with minimal accuracy loss (1.2% and 0.6% pass@1), and significantly boosts harmless rates on safety benchmarks (up to +21.7%). Our results highlight the potential of ASRR for enabling efficient, adaptive, and safer reasoning in LRMs.

Key facts. Cited by 26 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 28 of its 33 references that are in the map. First posted on arXiv 21 May 2025.

Builds on

28 of 33 references.

Closest ancestors · 12

Same direction, newest first

  1. 2025Dynamic Early Exit in Reasoning Models22 Apr
  2. 2025Thought Manipulation: External Thought Can Be Efficient for Large Reasoning Models18 Apr
  3. 2025Reasoning Models Can Be Effective Without Thinking14 Apr
  4. 2025Reasoning Models Know When They're Right: Probing Hidden States for Self-Verification7 Apr
  5. 2025Unlocking Efficient Long-to-Short LLM Reasoning with Model Merging26 Mar
  6. 2025DAST: Difficulty-Adaptive Slow-Thinking for Large Reasoning Models6 Mar
  7. 2025Revisiting the Test-Time Scaling of o1-like Models: Do they Truly Possess Test-Time Scaling Capabilities?17 Feb
  8. 2025s1: Simple test-time scaling31 Jan
  9. 2025Reward-Guided Speculative Decoding for Efficient LLM Reasoning31 Jan
  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. 2023Self-Evaluation Guided Beam Search for Reasoning1 May

Other references · 16

  1. 2017Proximal Policy Optimization Algorithmscited by 15,215
  2. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  3. 2023Let's Verify Step by Stepcited by 3,789
  4. 2024OlympiadBench: A Challenging Benchmark for Promoting AGI with Olympiad-Level Bilingual Multimodal Scientific Problemscited by 1,136
  5. 2025Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?cited by 982
  6. 2023BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Datasetcited by 860
  7. 2025Chain of Draft: Thinking Faster by Writing Lesscited by 297
  8. 2025Training Language Models to Reason Efficientlycited by 247
  9. 2023Red-Teaming Large Language Models using Chain of Utterances for Safety-Alignmentcited by 223
  10. 2025Optimizing Test-Time Compute via Meta Reinforcement Fine-Tuningcited by 123
  11. 2025Safety Tax: Safety Alignment Makes Your Large Reasoning Models Less Reasonablecited by 122
  12. 2025Sketch-of-Thought: Efficient LLM Reasoning with Adaptive Cognitive-Inspired Sketchingcited by 117
4 more
  1. 2025SoftCoT: Soft Chain-of-Thought for Efficient Reasoning with LLMscited by 116
  2. 2025OverThink: Slowdown Attacks on Reasoning LLMscited by 96
  3. 2024Learning to Route LLMs with Confidence Tokenscited by 67
  4. 2024Unlocking the Capabilities of Thought: A Reasoning Boundary Framework to Quantify and Optimize Chain-of-Thoughtcited by 56

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 13; 2026: 13.

26 papers.

Sort:

By area:

  1. 2025Does Thinking More always Help? Mirage of Test-Time Scaling in Reasoning Modelscited by 53
  2. 2025Reinforcement Learning Meets Large Language Models: A Survey of Advancements and Applications Across the LLM Lifecyclecited by 35
  3. 2026Dynamic Model Routing and Cascading for Efficient LLM Inference: A Surveycited by 19
  4. 2025Towards Concise and Adaptive Thinking in Large Reasoning Models: A Surveycited by 18
  5. 2025CodeDance: A Dynamic Tool-integrated MLLM for Executable Visual Reasoningcited by 16
  6. 2025Revisiting Entropy Regularization: Adaptive Coefficient Unlocks Its Potential for LLM Reinforcement Learningcited by 11
  7. 2026Sparse-RL: Breaking the Memory Wall in LLM Reinforcement Learning via Stable Sparse Rolloutscited by 9
  8. 2025Hierarchical Budget Policy Optimization for Adaptive Reasoningcited by 8
  9. 2025Thinking Sparks!: Emergent Attention Heads in Reasoning Models During Post Trainingcited by 8
  10. 2025Test-Time Scaling in Diffusion LLMs via Hidden Semi-Autoregressive Expertscited by 8
  11. 2026Neural Chain-of-Thought Search: Searching the Optimal Reasoning Path to Enhance Large Language Modelscited by 8
  12. 2026FoE: Forest of Errors Makes the First Solution the Best in Large Reasoning Modelscited by 8
14 more
  1. 2025PaTaRM: Bridging Pairwise and Pointwise Signals via Preference-Aware Task-Adaptive Reward Modelingcited by 6
  2. 2025Thinking-Free Policy Initialization Makes Distilled Reasoning Models More Effective and Efficient Reasonerscited by 5
  3. 2025DART: Difficulty-Adaptive Reasoning Truncation for Efficient Large Language Modelscited by 5
  4. 2025Overthinking Reduction with Decoupled Rewards and Curriculum Data Schedulingcited by 4
  5. 2026Beyond Speedup -- Utilizing KV Cache for Sampling and Reasoningcited by 3
  6. 2026CODA: Difficulty-Aware Compute Allocation for Adaptive Reasoningcited by 3
  7. 2025Adaptive Dual Reasoner: Large Reasoning Models Can Think Efficiently by Hybrid Reasoningcited by 2
  8. 2026PAEC: Position-Aware Entropy Calibration for LLM Reasoning in RLVRcited by 2
  9. 2026VisRef: Visual Refocusing while Thinking Improves Test-Time Scaling in Multi-Modal Large Reasoning Modelscited by 1
  10. 2026Balancing the Reasoning Load: Difficulty-Differentiated Policy Optimization with Length Redistribution for Efficient and Robust Reinforcement Learningcited by 0
  11. 2026Post-Reasoning: Improving the Performance of Non-Thinking Models at No Costcited by 0
  12. 2026Cluster, Route, Escalate: Cascaded Framework for Cost-Aware LLM Servingcited by 0
  13. 2026ThinkRetrieve: Retrieval-Augmented Reasoning Traces for Test-Time Scalingcited by 0
  14. 2026Construting Reverse Thinking: Developing Large Language Models' Reverse Thingking Abilitycited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Think or Not? Exploring Thinking Efficiency in Large Reasoning Models via an Information-Theoretic Lens
  2. 2025Trade-offs in Large Reasoning Models: An Empirical Analysis of Deliberative and Adaptive Reasoning over Foundational Capabilities
  3. 2026APR: Penalizing Structural Redundancy in Large Reasoning Models via Anchor-based Process Rewards
  4. 2026Stable Adaptive Thinking via Advantage Shaping and Length-Aware Gradient Regulation
  5. 2025OThink-R1: Intrinsic Fast/Slow Thinking Mode Switching for Over-Reasoning Mitigation
  6. 2025Optimizing Length Compression in Large Reasoning Models
  7. 2026CAT: Confidence-Adaptive Thinking for Efficient Reasoning of Large Reasoning Models
  8. 2025Exploring and Exploiting the Inherent Efficiency within Large Reasoning Models for Self-Guided Efficiency Enhancement