mlsanityData to 14 Oct 2026

Thinkless: LLM Learns When to Think

arXiv 2505.13379 ↗First posted 19 May 2025cs.CL cs.AIcited by 113

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

Abstract (arXiv)

Reasoning Language Models, capable of extended chain-of-thought reasoning, have demonstrated remarkable performance on tasks requiring complex logical inference. However, applying elaborate reasoning for all queries often results in substantial computational inefficiencies, particularly when many problems admit straightforward solutions. This motivates an open question: Can LLMs learn when to think? To answer this, we propose Thinkless, a learnable framework that empowers an LLM to adaptively select between short-form and long-form reasoning, based on both task complexity and the model's ability. Thinkless is trained under a reinforcement learning paradigm and employs two control tokens, <short> for concise responses and <think> for detailed reasoning. At the core of our method is a Decoupled Group Relative Policy Optimization (DeGRPO) algorithm, which decomposes the learning objective of hybrid reasoning into two components: (1) a control token loss that governs the selection of the reasoning mode, and (2) a response loss that improves the accuracy of the generated answers. This decoupled formulation enables fine-grained control over the contributions of each objective, stabilizing training and effectively preventing collapse observed in vanilla GRPO. Empirically, on several benchmarks such as Minerva Algebra, MATH-500, and GSM8K, Thinkless is able to reduce the usage of long-chain thinking by 50% - 90%, significantly improving the efficiency of Reasoning Language Models. The code is available at https://github.com/VainF/Thinkless

Key facts. Cited by 113 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 33 of its 45 references that are in the map. First posted on arXiv 19 May 2025.

Builds on

33 of 45 references.

Closest ancestors · 10

Same direction, newest first

  1. 2025Efficient Reasoning Models: A Survey15 Apr
  2. 2025L1: Controlling How Long A Reasoning Model Thinks With Reinforcement Learning6 Mar
  3. 2025Reasoning with Latent Thoughts: On the Power of Looped Transformers24 Feb
  4. 2025CoT-Valve: Length-Compressible Chain-of-Thought Tuning13 Feb
  5. 2025The Danger of Overthinking: Examining the Reasoning-Action Dilemma in Agentic Tasks12 Feb
  6. 2025Reward-Guided Speculative Decoding for Efficient LLM Reasoning31 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

Other references · 23

  1. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,459
  2. 2015Distilling the Knowledge in a Neural Networkcited by 10,324
  3. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  4. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,379
  5. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  6. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,371
  7. 2024Qwen2.5 Technical Reportcited by 4,942
  8. 2023Let's Verify Step by Stepcited by 3,789
  9. 2024HybridFlow: A Flexible and Efficient RLHF Frameworkcited by 2,057
  10. 2019Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelismcited by 1,868
  11. 2025Understanding R1-Zero-Like Training: A Critical Perspectivecited by 1,345
  12. 2025Kimi k1.5: Scaling Reinforcement Learning with LLMscited by 1,210
11 more
  1. 2024RouteLLM: Learning to Route LLMs with Preference Datacited by 682
  2. 2025SimpleRL-Zoo: Investigating and Taming Zero Reinforcement Learning for Open Base Models in the Wildcited by 511
  3. 2025Chain of Draft: Thinking Faster by Writing Lesscited by 297
  4. 2022Teaching Small Language Models to Reasoncited by 284
  5. 2025ReSearch: Learning to Reason with Search for LLMs via Reinforcement Learningcited by 161
  6. 2024Distilling System 2 into System 1cited by 159
  7. 2025Sketch-of-Thought: Efficient LLM Reasoning with Adaptive Cognitive-Inspired Sketchingcited by 117
  8. 2025Small Models Struggle to Learn from Strong Reasonerscited by 113
  9. 2025Towards Reasoning Ability of Small Language Modelscited by 29
  10. 2023Mixed Distillation Helps Smaller Language Model Better Reasoningcited by 19
  11. 2024Distilling Reasoning Ability from Large Language Models with Adaptive Thinkingcited by 13

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 49; 2026: 64.

113 papers.

Sort:

By area:

  1. 2025AdaThinkDrive: Adaptive Thinking via Reinforcement Learning for Autonomous Drivingcited by 58
  2. 2025Does Thinking More always Help? Mirage of Test-Time Scaling in Reasoning Modelscited by 53
  3. 2025Implicit Reasoning in Large Language Models: A Comprehensive Surveycited by 43
  4. 2025Don't Overthink It: A Survey of Efficient R1-style Large Reasoning Modelscited by 37
  5. 2025Reinforcement Learning Meets Large Language Models: A Survey of Advancements and Applications Across the LLM Lifecyclecited by 35
  6. 2025DLER: Doing Length pEnalty Right - Incentivizing More Intelligence per Token via Reinforcement Learningcited by 35
  7. 2025A Survey of Token Compression for Efficient Multimodal Large Language Modelscited by 25
  8. 2025VeriThinker: Learning to Verify Makes Reasoning Model Efficientcited by 24
  9. 2025Mini-Omni-Reasoner: Token-Level Thinking-in-Speaking in Large Speech Modelscited by 21
  10. 2025Train Long, Think Short: Curriculum Learning for Efficient Reasoningcited by 20
  11. 2025OBS-Diff: Accurate Pruning For Diffusion Models in One-Shotcited by 19
  12. 2025Towards Concise and Adaptive Thinking in Large Reasoning Models: A Surveycited by 18
38 more
  1. 2025DRPO: Efficient Reasoning via Decoupled Reward Policy Optimizationcited by 16
  2. 2025Do Thinking Tokens Help or Trap? Towards More Efficient Large Reasoning Modelcited by 14
  3. 2025Controlling Performance and Budget of a Centralized Multi-agent LLM System with Reinforcement Learningcited by 13
  4. 2025AutoL2S: Auto Long-Short Reasoning for Efficient Large Language Modelscited by 12
  5. 2025KAT-V1: Kwai-AutoThink Technical Reportcited by 12
  6. 2025Retrieval-of-Thought: Efficient Reasoning via Reusing Thoughtscited by 12
  7. 2025Position: The Hidden Costs and Measurement Gaps of Reinforcement Learning with Verifiable Rewardscited by 12
  8. 2025LAPO: Internalizing Reasoning Efficiency via Length-Adaptive Policy Optimizationcited by 11
  9. 2025DiffAdapt: Difficulty-Adaptive Reasoning for Token-Efficient LLM Inferencecited by 11
  10. 2025ConciseHint: Boosting Efficient Reasoning via Continuous Concise Hints during Generationcited by 10
  11. 2025Do LLMs Really Need 10+ Thoughts for "Find the Time 1000 Days Later"? Towards Structural Understanding of LLM Overthinkingcited by 10
  12. 2025How Far Are We from Optimal Reasoning Efficiency?cited by 8
  13. 2025Hierarchical Budget Policy Optimization for Adaptive Reasoningcited by 8
  14. 2025Test-Time Scaling in Diffusion LLMs via Hidden Semi-Autoregressive Expertscited by 8
  15. 2025Probing the Difficulty Perception Mechanism of Large Language Modelscited by 8
  16. 2025R-Horizon: How Far Can Your Large Reasoning Model Really Go in Breadth and Depth?cited by 8
  17. 2025Schoenfeld's Anatomy of Mathematical Reasoning by Language Modelscited by 8
  18. 2026Neural Chain-of-Thought Search: Searching the Optimal Reasoning Path to Enhance Large Language Modelscited by 8
  19. 2026Fake-HR1: Rethinking Reasoning of Vision Language Model for Synthetic Image Detectioncited by 8
  20. 2026ThinkRouter: Efficient Reasoning via Routing Thinking between Latent and Discrete Spacescited by 8
  21. 2026Same Signal, Opposite Meaning: Direction-Informed Adaptive Learning for LLM Agentscited by 8
  22. 2025SmartThinker: Learning to Compress and Preserve Reasoning by Step-Level Length Controlcited by 7
  23. 2025R-Stitch: Dynamic Trajectory Stitching for Efficient Reasoningcited by 7
  24. 2026ORBIT: On-policy Exploration-Exploitation for Controllable Multi-Budget Reasoningcited by 7
  25. 2025On the Provable Performance Guarantee of Efficient Reasoning Modelscited by 6
  26. 2025Asking like Socrates: Socrates helps VLMs understand remote sensing imagescited by 6
  27. 2026Mid-Think: Training-Free Intermediate-Budget Reasoning via Token-Level Triggerscited by 6
  28. 2026Towards Efficient Large Language Reasoning Models via Extreme-Ratio Chain-of-Thought Compressioncited by 6
  29. 2026Mitigating Overthinking in Large Reasoning Language Models via Reasoning Path Deviation Monitoringcited by 6
  30. 2026When Is Thinking Enough? Early Exit via Sufficiency Assessment for Efficient Reasoningcited by 6
  31. 2025Think How to Think: Mitigating Overthinking with Autonomous Difficulty Cognition in Large Reasoning Modelscited by 5
  32. 2025StyleBench: Evaluating thinking styles in Large Language Modelscited by 5
  33. 2025Thinking-Free Policy Initialization Makes Distilled Reasoning Models More Effective and Efficient Reasonerscited by 5
  34. 2026DF-RAG: Query-Aware Diversity for Retrieval-Augmented Generationcited by 5
  35. 2026FROST: Filtering Reasoning Outliers with Attention for Efficient Reasoningcited by 5
  36. 2026From Reasoning Traces to Reusable Modules: Understanding Compositional Generalization in Language Model Reasoningcited by 5
  37. 2025From "Aha Moments" to Controllable Thinking: Toward Meta-Cognitive Reasoning in Large Reasoning Models via Decoupled Reasoning and Controlcited by 4
  38. 2025The Zero-Step Thinking: An Empirical Study of Mode Selection as Harder Early Exit in Reasoning Modelscited by 4

50 most cited of 113 All 113 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2025Not All Thoughts are Generated Equal: Efficient LLM Reasoning via Multi-Turn Reinforcement Learning
  2. 2025ThinkPrune: Pruning Long Chain-of-Thought of LLMs via Reinforcement Learning
  3. 2025Thinking on the Fly: Test-Time Reasoning Enhancement via Latent Thought Policy Optimization
  4. 2025RL of Thoughts: Navigating LLM Reasoning with Inference-time Reinforcement Learning
  5. 2026Policy of Thoughts: Scaling Test-Time Training for LLM Reasoning via Online Policy Evolution
  6. 2025Think or Not? Selective Reasoning via Reinforcement Learning for Vision-Language Models
  7. 2026Knowing When Thinking Is Not Enough: Teaching Small Reasoning Models to Reason Beyond Their Parametric Knowledge
  8. 2025The Markovian Thinker: Architecture-Agnostic Linear Scaling of Reasoning