mlsanityData to 14 Oct 2026

Not All Tokens Are What You Need In Thinking

arXiv 2505.17827 ↗First posted 23 May 2025cs.CLcited by 20

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

Abstract (arXiv)

Modern reasoning models, such as OpenAI's o1 and DeepSeek-R1, exhibit impressive problem-solving capabilities but suffer from critical inefficiencies: high inference latency, excessive computational resource consumption, and a tendency toward overthinking -- generating verbose chains of thought (CoT) laden with redundant tokens that contribute minimally to the final answer. To address these issues, we propose Conditional Token Selection (CTS), a token-level compression framework with a flexible and variable compression ratio that identifies and preserves only the most essential tokens in CoT. CTS evaluates each token's contribution to deriving correct answers using conditional importance scoring, then trains models on compressed CoT. Extensive experiments demonstrate that CTS effectively compresses long CoT while maintaining strong reasoning performance. Notably, on the GPQA benchmark, Qwen2.5-14B-Instruct trained with CTS achieves a 9.1% accuracy improvement with 13.2% fewer reasoning tokens (13% training token reduction). Further reducing training tokens by 42% incurs only a marginal 5% accuracy drop while yielding a 75.8% reduction in reasoning tokens, highlighting the prevalence of redundancy in existing CoT.

Key facts. Cited by 20 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 21 of its 32 references that are in the map. First posted on arXiv 23 May 2025.

Builds on

21 of 32 references.

Closest ancestors · 9

Same direction, newest first

  1. 2025Reasoning Models Can Be Effective Without Thinking14 Apr
  2. 2025Effectively Controlling Reasoning Models through Thinking Intervention31 Mar
  3. 2025A Survey on Test-Time Scaling in Large Language Models: What, How, Where, and How Well?31 Mar
  4. 2025Think Twice: Enhancing LLM Reasoning by Scaling Multi-round Test-time Thinking25 Mar
  5. 2025Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models20 Mar
  6. 2025How Well do LLMs Compress Their Own Chain-of-Thought? A Token Complexity Approach3 Mar
  7. 2025Demystifying Long Chain-of-Thought Reasoning in LLMs5 Feb
  8. 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs30 Dec
  9. 2024C3oT: Generating Shorter Chain-of-Thought without Compromising Effectiveness16 Dec

Other references · 12

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2020Retrieval-Augmented Generation for Knowledge-Intensive NLP Taskscited by 9,297
  3. 2020Scaling Laws for Neural Language Modelscited by 6,586
  4. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,371
  5. 2022Large Language Models are Zero-Shot Reasonerscited by 4,374
  6. 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,879
  7. 2025Kimi k1.5: Scaling Reinforcement Learning with LLMscited by 1,210
  8. 2023LongLLMLingua: Accelerating and Enhancing LLMs in Long Context Scenarios via Prompt Compressioncited by 453
  9. 2024LLMLingua-2: Data Distillation for Efficient and Faithful Task-Agnostic Prompt Compressioncited by 348
  10. 2025Chain of Draft: Thinking Faster by Writing Lesscited by 297
  11. 2025AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning datasetcited by 174
  12. 20251.4 Million Open-Source Distilled Reasoning Dataset to Empower Large Language Model Trainingcited by 49

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 5; 2026: 15.

20 papers.

Sort:

By area:

  1. 2025Don't Overthink It: A Survey of Efficient R1-style Large Reasoning Modelscited by 37
  2. 2025DRPO: Efficient Reasoning via Decoupled Reward Policy Optimizationcited by 16
  3. 2026Replaying pre-training data improves fine-tuningcited by 13
  4. 2026Neural Chain-of-Thought Search: Searching the Optimal Reasoning Path to Enhance Large Language Modelscited by 8
  5. 2026Towards Efficient Large Language Reasoning Models via Extreme-Ratio Chain-of-Thought Compressioncited by 6
  6. 2025PEAR: Phase Entropy Aware Reward for Efficient Reasoningcited by 5
  7. 2026Steering Large Reasoning Models towards Concise Reasoning via Flow Matchingcited by 4
  8. 2026Efficient Reasoning with Balanced Thinkingcited by 4
  9. 2025ThinkPilot: Steering Reasoning Models via Automated Think-prefixes Optimizationcited by 3
  10. 2026Draft-Thinking: Learning Efficient Reasoning in Long Chain-of-Thought LLMscited by 3
  11. 2026Learning from Partial Chain-of-Thought via Truncated-Reasoning Self-Distillationcited by 3
  12. 2026Shorthand for Thought: Compressing LLM Reasoning via Entropy-Guided Supertokenscited by 2
8 more
  1. 2025Focused Chain-of-Thought: Efficient LLM Reasoning via Structured Input Informationcited by 1
  2. 2026A State-Transition Framework for Efficient LLM Reasoningcited by 1
  3. 2026Reasoning Denoiser: Denoising Reasoning Traces for Hallucination Detection in Large Reasoning Modelscited by 1
  4. 2026Bridging Efficiency and Transparency: Explainable CoT Compression in Multimodal Large Reasoning Modelscited by 0
  5. 2026SkillRater: Untangling Capabilities in Multimodal Datacited by 0
  6. 2026Stateful Reasoning via Insight Replaycited by 0
  7. 2026Revisiting Complete Reasoning Traces for Post-Trainingcited by 0
  8. 2026Efficient Reasoning via Constrained Optimization in Latent Spacecited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Every Token Leaves a Ripple in the Stream of Thought: Eliciting Model-Internal Token Saliency for Chain-of-Thought Compression
  2. 2025TokenSqueeze: Performance-Preserving Compression for Reasoning LLMs
  3. 2025TokenSkip: Controllable Chain-of-Thought Compression in LLMs
  4. 2025Less Is More: Fast and Accurate Reasoning with Cross-Head Unified Sparse Attention
  5. 2026DoubtLess: Training-Free Contrastive Decoding for Efficient Reasoning
  6. 2025Optimizing Length Compression in Large Reasoning Models
  7. 2026On the Token Value Inequality in Efficient Reasoning
  8. 2026Quantization Inflates Reasoning: Token Inflation as a Hidden Cost of Low-Bit Reasoning Models