mlsanityData to 14 Oct 2026

ThinkLess: A Training-Free Inference-Efficient Method for Reducing Reasoning Redundancy

arXiv 2505.15684 ↗First posted 21 May 2025cs.CLcited by 17

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

also in 1 more
Abstract (arXiv)

While Chain-of-Thought (CoT) prompting improves reasoning in large language models (LLMs), the excessive length of reasoning tokens increases latency and KV cache memory usage, and may even truncate final answers under context limits. We propose ThinkLess, an inference-efficient framework that terminates reasoning generation early and maintains output quality without modifying the model. Atttention analysis reveals that answer tokens focus minimally on earlier reasoning steps and primarily attend to the reasoning terminator token, due to information migration under causal masking. Building on this insight, ThinkLess inserts the terminator token at earlier positions to skip redundant reasoning while preserving the underlying knowledge transfer. To prevent format discruption casued by early termination, ThinkLess employs a lightweight post-regulation mechanism, relying on the model's natural instruction-following ability to produce well-structured answers. Without fine-tuning or auxiliary data, ThinkLess achieves comparable accuracy to full-length CoT decoding while greatly reducing decoding time and memory consumption.

Key facts. Cited by 17 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 40 references that are in the map. First posted on arXiv 21 May 2025.

Builds on

28 of 40 references.

Closest ancestors · 10

Same direction, newest first

  1. 2025A Survey of Efficient Reasoning for Large Reasoning Models: Language, Multimodality, and Beyond27 Mar
  2. 2025Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models20 Mar
  3. 2025DAST: Difficulty-Adaptive Slow-Thinking for Large Reasoning Models6 Mar
  4. 2025LightThinker: Thinking Step-by-Step Compression21 Feb
  5. 2025TokenSkip: Controllable Chain-of-Thought Compression in LLMs17 Feb
  6. 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs30 Dec
  7. 2024Token-Budget-Aware LLM Reasoning24 Dec
  8. 2024Forest-of-Thought: Scaling Test-Time Compute for Enhancing LLM Reasoning12 Dec
  9. 2024A Comparative Study on Reasoning Patterns of OpenAI's o1 Model17 Oct
  10. 2024Self-Contrast: Better Reflection Through Inconsistent Solving Perspectives4 Jan

Other references · 18

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  3. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  4. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  5. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,708
  6. 2024Qwen2.5 Technical Reportcited by 4,942
  7. 2022Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Themcited by 1,807
  8. 2025Understanding R1-Zero-Like Training: A Critical Perspectivecited by 1,345
  9. 2024OpenAI o1 System Cardcited by 1,244
  10. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 797
  11. 2022Automatic Chain of Thought Prompting in Large Language Modelscited by 688
  12. 2025Chain of Draft: Thinking Faster by Writing Lesscited by 297
6 more
  1. 2022Reasoning with Language Model Prompting: A Surveycited by 252
  2. 2025Optimizing Test-Time Compute via Meta Reinforcement Fine-Tuningcited by 123
  3. 2025Sketch-of-Thought: Efficient LLM Reasoning with Adaptive Cognitive-Inspired Sketchingcited by 117
  4. 2025CAKE: Cascading and Adaptive KV Cache Eviction with Layer Preferencescited by 58
  5. 2024Multi-Agent Sampling: Scaling Inference Compute for Data Synthesis with Tree Search-Based Agentic Collaborationcited by 9
  6. 2025Will Pre-Training Ever End? A First Step Toward Next-Generation Foundation MLLMs via Self-Improving Systematic Cognitioncited by 7

Built on by

Cited by year (2026 to 14 Oct)

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

17 papers.

Sort:

By area:

  1. 2025Critique-GRPO: Advancing LLM Reasoning with Natural Language and Numerical Feedbackcited by 90
  2. 2025XRPO: Pushing the limits of GRPO with Targeted Exploration and Exploitationcited by 18
  3. 2025Are Large Reasoning Models Interruptible?cited by 7
  4. 2025Asynchronous Reasoning: Training-Free Interactive Thinking LLMscited by 7
  5. 2026Mitigating Overthinking in Large Reasoning Language Models via Reasoning Path Deviation Monitoringcited by 6
  6. 2026Seeing Through the Chain: Mitigate Hallucination in Multimodal Reasoning Models via CoT Compression and Contrastive Preference Optimizationcited by 5
  7. 2025The Zero-Step Thinking: An Empirical Study of Mode Selection as Harder Early Exit in Reasoning Modelscited by 4
  8. 2026BFS-PO: Best-First Search for Large Reasoning Modelscited by 2
  9. 2025ThinkBrake: Efficient Reasoning via Log-Probability Margin Guided Decodingcited by 1
  10. 2025Inference-Time Chain-of-Thought Pruning with Latent Informativeness Signalscited by 1
  11. 2025ChainV: Atomic Visual Hints Make Multimodal Reasoning Shorter and Bettercited by 1
  12. 2026CRISP: Compressing Redundancy in Chain-of-Thought via Intrinsic Saliency Pruningcited by 1
5 more
  1. 2025Strategic Reflectivism In Intelligent Systemscited by 0
  2. 2025Dual-Process Scaffold Reasoning for Enhancing LLM Code Debuggingcited by 0
  3. 2026Multi-objective Evolutionary Merging Enables Efficient Reasoning Modelscited by 0
  4. 2026Post-Reasoning: Improving the Performance of Non-Thinking Models at No Costcited by 0
  5. 2026DRT: Dense Reasoning Trace for Efficient and Grounded Multimodal Reasoningcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026SyncThink: A Training-Free Strategy to Align Inference Termination with Reasoning Saturation
  2. 2026DoubtLess: Training-Free Contrastive Decoding for Efficient Reasoning
  3. 2026OS-Pruner: Pruning Chains-of-Thought of Reasoning Models via Optimal Stopping
  4. 2026</think> Doesn't Stop Reasoning: Analysis of Spurious CoT Termination
  5. 2025Efficient Reasoning via Thought-Training and Thought-Free Inference
  6. 2025Can Reasoning Path still be Effective as Input? Bridging Post-Reasoning to Chain-of-Thought Compression
  7. 2025Early Stopping Chain-of-thoughts in Large Language Models
  8. 2026TERMINATOR: Learning Optimal Exit Points for Early Stopping in Chain-of-Thought Reasoning