mlsanityData to 14 Oct 2026

Can Pruning Improve Reasoning? Revisiting Long-CoT Compression with Capability in Mind for Better Reasoning

arXiv 2505.14582 ↗First posted 20 May 2025cs.CLcited by 19

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

Abstract (arXiv)

Long chain-of-thought (Long-CoT) reasoning improves accuracy in LLMs, yet its verbose, self-reflective style often hinders effective distillation into small language models (SLMs). We revisit Long-CoT compression through the lens of capability alignment and ask: Can pruning improve reasoning? We propose Prune-on-Logic, a structure-aware framework that transforms Long-CoT into logic graphs and selectively prunes low-utility reasoning steps under self-verification constraints. Through systematic analysis across three pruning strategies targeting entire chains, core reasoning, and verification, we find that verification pruning consistently improves accuracy while reducing token usage, whereas pruning reasoning steps or indiscriminate pruning degrades performance. Our study reveals that effective pruning aligns supervision with model capacity rather than merely shortening inputs. Gains hold across tasks, model scales, and CoT capability, with larger models benefiting more from pruning due to richer but more redundant reasoning. Our empirical findings highlight pruning as a structural optimization strategy for aligning CoT reasoning with SLM capacity.

Key facts. Cited by 19 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 23 of its 41 references that are in the map. First posted on arXiv 20 May 2025.

Builds on

23 of 41 references.

Closest ancestors · 3

Same direction, newest first

  1. 2025LightThinker: Thinking Step-by-Step Compression21 Feb
  2. 2025Demystifying Long Chain-of-Thought Reasoning in LLMs5 Feb
  3. 2025O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruning22 Jan

Other references · 20

  1. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  2. 2023Let's Verify Step by Stepcited by 3,789
  3. 2018CommonsenseQA: A Question Answering Challenge Targeting Commonsense Knowledgecited by 1,822
  4. 2022Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Themcited by 1,807
  5. 2025LIMO: Less is More for Reasoningcited by 498
  6. 2025Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Modelscited by 384
  7. 2024SWIFT:A Scalable lightWeight Infrastructure for Fine-Tuningcited by 355
  8. 2024LLMLingua-2: Data Distillation for Efficient and Faithful Task-Agnostic Prompt Compressioncited by 348
  9. 2024When Scaling Meets LLM Finetuning: The Effect of Data, Model and Finetuning Methodcited by 204
  10. 2022Distilling Reasoning Capabilities into Smaller Language Modelscited by 158
  11. 2025Small Models Struggle to Learn from Strong Reasonerscited by 113
  12. 2025PRISM: Self-Pruning Intrinsic Selection Method for Training-Free Multimodal Data Selectioncited by 61
8 more
  1. 2024Beyond Examples: High-level Automated Reasoning Paradigm in In-Context Learning via MCTScited by 54
  2. 2025Deconstructing Long Chain-of-Thought: A Structured Reasoning Optimization Framework for Long CoT Distillationcited by 43
  3. 2024RATT: A Thought Structure for Coherent and Correct LLM Reasoningcited by 41
  4. 2024CoMT: Chain-of-Medical-Thought Reduces Hallucination in Medical Report Generationcited by 25
  5. 2024Markov Chain of Thought for Efficient Mathematical Reasoningcited by 19
  6. 2025Adaptive Graph of Thoughts: Test-Time Adaptive Reasoning Unifying Chain, Tree, and Graph Structurescited by 16
  7. 2024Learn Beyond The Answer: Training Language Models with Reflection for Mathematical Reasoningcited by 15
  8. 2025Self-Correction Makes LLMs Better Parserscited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 9; 2026: 10.

19 papers.

Sort:

By area:

  1. 2025A Survey of Context Engineering for Large Language Modelscited by 126
  2. 2025Reducing Cost of LLM Agents with Trajectory Reductioncited by 48
  3. 2025Don't Overthink It: A Survey of Efficient R1-style Large Reasoning Modelscited by 37
  4. 2025Every Rollout Counts: Optimal Resource Allocation for Efficient Test-Time Scalingcited by 15
  5. 2026Neural Chain-of-Thought Search: Searching the Optimal Reasoning Path to Enhance Large Language Modelscited by 8
  6. 2025PEAR: Phase Entropy Aware Reward for Efficient Reasoningcited by 5
  7. 2026Characterizing, Evaluating, and Optimizing Complex Reasoningcited by 5
  8. 2025SDEval: Safety Dynamic Evaluation for Multimodal Large Language Modelscited by 3
  9. 2025SatireDecoder: Visual Cascaded Decoupling for Enhancing Satirical Image Comprehensioncited by 3
  10. 2025Reinforcement Learning for Chain of Thought Compression with One-Domain-to-All Generalizationcited by 3
  11. 2025Chopping Trees: Semantic Similarity Based Dynamic Pruning for Tree-of-Thought Reasoningcited by 1
  12. 2026STOP: Structured On-Policy Pruning of Long-Form Reasoning in Low-Data Regimescited by 1
7 more
  1. 2026Mapping Text to Multiplex Graph: Prompt Compression as L'evy Walk-Guided Graph Pruningcited by 1
  2. 2026Bridging Efficiency and Transparency: Explainable CoT Compression in Multimodal Large Reasoning Modelscited by 0
  3. 2026LegalDrill: Diagnosis-Driven Synthesis for Legal Reasoning in Small Language Modelscited by 0
  4. 2026Breaking the Reward Barrier: Accelerating Tree-of-Thought Reasoning via Speculative Explorationcited by 0
  5. 2026When Compression Helps and When It Hurts: Condition-Aware Analysis of Chain-of-Thought Distillationcited by 0
  6. 2026Can Reasoning Models Detect Changes to their Chains of Thought?cited by 0
  7. 2026VIG: Visual Information Gain as a Reward Signal for Multimodal Chain-of-Thought Compressioncited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Pruning the Unsurprising: Efficient LLM Reasoning via First-Token Surprisal
  2. 2026Graph-Based Chain-of-Thought Pruning for Reducing Redundant Reflections in Reasoning LLMs
  3. 2025Think Before You Prune: Self-Reflective Structured Pruning for Reasoning Language Models
  4. 2026From LLMs to LRMs: Rethinking Pruning for Reasoning-Centric Models
  5. 2026Revisiting the Effectiveness of LLM Pruning for Test-Time Scaling
  6. 2025When Fewer Layers Break More Chains: Layer Pruning Harms Test-Time Scaling in LLMs
  7. 2026CtrlCoT: Dual-Granularity Chain-of-Thought Compression for Controllable Reasoning
  8. 2026OS-Pruner: Pruning Chains-of-Thought of Reasoning Models via Optimal Stopping