mlsanityData to 14 Oct 2026

TL;DR: Too Long, Do Re-weighting for Efficient LLM Reasoning Compression

arXiv 2506.02678 ↗First posted 3 Jun 2025cs.CL cs.CE cs.NAcited by 16

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

Abstract (arXiv)

Large Language Models (LLMs) have recently achieved remarkable progress by leveraging Reinforcement Learning and extended Chain-of-Thought (CoT) techniques. However, the challenge of performing efficient language reasoning--especially during inference with extremely long outputs--has drawn increasing attention from the research community. In this work, we propose a dynamic ratio-based training pipeline that does not rely on sophisticated data annotations or interpolation between multiple models. We continuously balance the weights between the model's System-1 and System-2 data to eliminate redundant reasoning processes while preserving the model's reasoning capability. We validate our approach across models on DeepSeek-R1-Distill-7B and DeepSeek-R1-Distill-14B and on a diverse set of benchmarks with varying difficulty levels. Our method significantly reduces the number of output tokens by nearly 40% while maintaining the accuracy of the reasoning. Our code and data will be available soon.

Key facts. Cited by 16 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 22 of its 58 references that are in the map. First posted on arXiv 3 Jun 2025.

Builds on

22 of 58 references.

Closest ancestors · 8

Same direction, newest first

  1. 2025Activation-Guided Consensus Merging for Large Language Models20 May
  2. 2025L1: Controlling How Long A Reasoning Model Thinks With Reinforcement Learning6 Mar
  3. 2025TokenSkip: Controllable Chain-of-Thought Compression in LLMs17 Feb
  4. 2025Efficient Reasoning with Hidden Thinking31 Jan
  5. 2025O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruning22 Jan
  6. 2024Token-Budget-Aware LLM Reasoning24 Dec
  7. 2024Compressed Chain of Thought: Efficient Reasoning Through Dense Representations17 Dec
  8. 2024C3oT: Generating Shorter Chain-of-Thought without Compromising Effectiveness16 Dec

Other references · 14

  1. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  2. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,371
  3. 2025Kimi k1.5: Scaling Reinforcement Learning with LLMscited by 1,210
  4. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 797
  5. 2021A Diverse Corpus for Evaluating and Developing English Math Word Problem Solverscited by 318
  6. 2025Chain of Draft: Thinking Faster by Writing Lesscited by 297
  7. 2025OverThink: Slowdown Attacks on Reasoning LLMscited by 96
  8. 2024Learning to Route LLMs with Confidence Tokenscited by 67
  9. 2025RedStar: Does Scaling Long-CoT Data Unlock Better Slow-Reasoning Systems?cited by 60
  10. 2025DRP: Distilled Reasoning Pruning with Skill-aware Step Decomposition for Efficient Large Reasoning Modelscited by 30
  11. 2025Chain-of-Reasoning: Towards Unified Mathematical Reasoning in Large Language Models via a Multi-Paradigm Perspectivecited by 29
  12. 2024BPO: Towards Balanced Preference Optimization between Knowledge Breadth and Depth in Alignmentcited by 20
2 more
  1. 2024Explore the Reasoning Capability of LLMs in the Chess Testbedcited by 19
  2. 2024EffiCoder: Enhancing Code Generation in Large Language Models through Efficiency-Aware Fine-tuningcited by 17

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 9; 2026: 7.

16 papers.

Sort:

By area:

  1. 2025The Landscape of Agentic Reinforcement Learning for LLMs: A Surveycited by 205
  2. 2025Beyond Pass@1: Self-Play with Variational Problem Synthesis Sustains RLVRcited by 54
  3. 2025SwS: Self-aware Weakness-driven Problem Synthesis in Reinforcement Learning for LLM Reasoningcited by 42
  4. 2025Semantic-Space Exploration and Exploitation in RLVR for LLM Reasoningcited by 17
  5. 2025Aware First, Think Less: Dynamic Boundary Self-Awareness Drives Extreme Reasoning Efficiency in Large Language Modelscited by 13
  6. 2025DiffAdapt: Difficulty-Adaptive Reasoning for Token-Efficient LLM Inferencecited by 11
  7. 2025Asymmetric Proximal Policy Optimization: mini-critics boost LLM reasoningcited by 10
  8. 2026Neural Chain-of-Thought Search: Searching the Optimal Reasoning Path to Enhance Large Language Modelscited by 8
  9. 2026VTC-R1: Vision-Text Compression for Efficient Long-Context Reasoningcited by 8
  10. 2026When Silence Is Golden: Can LLMs Learn to Abstain in Temporal QA and Beyond?cited by 7
  11. 2026Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Modelscited by 7
  12. 2026Accordion-Thinking: Self-Regulated Step Summaries for Efficient and Readable LLM Reasoningcited by 6
4 more
  1. 2025Fast, Slow, and Tool-augmented Thinking for LLMs: A Reviewcited by 2
  2. 2026ConMax: Confidence-Maximizing Compression for Efficient Chain-of-Thought Reasoningcited by 2
  3. 2025SCALE: Selective Resource Allocation for Overcoming Performance Bottlenecks in Mathematical Test-time Scalingcited by 1
  4. 2026Graph-Based Chain-of-Thought Pruning for Reducing Redundant Reflections in Reasoning LLMscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Towards Efficient Large Language Reasoning Models via Extreme-Ratio Chain-of-Thought Compression
  2. 2025Fast on the Easy, Deep on the Hard: Efficient Reasoning via Powered Length Penalty
  3. 2025Training Language Models to Reason Efficiently
  4. 2025Think Silently, Think Fast: Dynamic Latent Compression of LLM Reasoning Chains
  5. 2025Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models
  6. 2026The Art of Efficient Reasoning: Data, Reward, and Optimization
  7. 2026Zipping the Thought: When and How Compressed Reasoning Data Works in LLM Post-Training
  8. 2025Reasoning Beyond Limits: Advances and Open Problems for LLMs