mlsanityData to 7 Oct 2026

DeepCompress: A Dual Reward Strategy for Dynamically Exploring and Compressing Reasoning Chains

arXiv 2510.27419 ↗First posted 31 Oct 2025cs.AI cs.CLcited by 7

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

Abstract (arXiv)

Large Reasoning Models (LRMs) have demonstrated impressive capabilities but suffer from cognitive inefficiencies like "overthinking" simple problems and "underthinking" complex ones. While existing methods that use supervised fine-tuning (SFT) or reinforcement learning (RL) with token-length rewards can improve efficiency, they often do so at the cost of accuracy. This paper introduces DeepCompress, a novel framework that simultaneously enhances both the accuracy and efficiency of LRMs. We challenge the prevailing approach of consistently favoring shorter reasoning paths, showing that longer responses can contain a broader range of correct solutions for difficult problems. DeepCompress employs an adaptive length reward mechanism that dynamically classifies problems as "Simple" or "Hard" in real-time based on the model's evolving capability. It encourages shorter, more efficient reasoning for "Simple" problems while promoting longer, more exploratory thought chains for "Hard" problems. This dual-reward strategy enables the model to autonomously adjust its Chain-of-Thought (CoT) length, compressing reasoning for well-mastered problems and extending it for those it finds challenging. Experimental results on challenging mathematical benchmarks show that DeepCompress consistently outperforms baseline methods, achieving superior accuracy while significantly improving token efficiency.

Key facts. Cited by 7 papers in the map (data through 7 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 24 of its 40 references that are in the map. First posted on arXiv 31 Oct 2025.

Builds on

24 of 40 references.

Closest ancestors · 13

Same direction, newest first

  1. 2025Incentivizing Dual Process Thinking for Efficient Large Language Model Reasoning22 May
  2. 2025Learn to Reason Efficiently with Adaptive Length-based Reward Shaping21 May
  3. 2025Shorten After You're Right: Lazy Length Penalties for Reasoning RL18 May
  4. 2025ShorterBetter: Guiding Reasoning Models to Find Optimal Inference Length for Efficient Reasoning30 Apr
  5. 2025ThinkPrune: Pruning Long Chain-of-Thought of LLMs via Reinforcement Learning2 Apr
  6. 2025L1: Controlling How Long A Reasoning Model Thinks With Reinforcement Learning6 Mar
  7. 2025Towards Thinking-Optimal Scaling of Test-Time Compute for LLM Reasoning25 Feb
  8. 2025Thoughts Are All Over the Place: On the Underthinking of o1-Like LLMs30 Jan
  9. 2025Think Smarter not Harder: Adaptive Reasoning with Inference Aware Optimization29 Jan
  10. 2025O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruning22 Jan
  11. 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs30 Dec
  12. 2024Token-Budget-Aware LLM Reasoning24 Dec
1 more
  1. 2024Concise Thoughts: Impact of Output Length on LLM Reasoning and Cost29 Jul

Other references · 11

  1. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  2. 2021Evaluating Large Language Models Trained on Codecited by 8,053
  3. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  4. 2025DAPO: An Open-Source LLM Reinforcement Learning System at Scalecited by 2,678
  5. 2025Kimi k1.5: Scaling Reinforcement Learning with LLMscited by 1,193
  6. 2025SimpleRL-Zoo: Investigating and Taming Zero Reinforcement Learning for Open Base Models in the Wildcited by 505
  7. 2025Cognitive Behaviors that Enable Self-Improving Reasoners, or, Four Habits of Highly Effective STaRscited by 373
  8. 2025DeepMath-103K: A Large-Scale, Challenging, Decontaminated, and Verifiable Mathematical Dataset for Advancing Reasoningcited by 261
  9. 2025Training Language Models to Reason Efficientlycited by 242
  10. 2025PolyMath: Evaluating Mathematical Reasoning in Multilingual Contextscited by 52
  11. 2025REA-RL: Reflection-Aware Online Reinforcement Learning for Efficient Reasoningcited by 5

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 7.

7 papers.

  1. 2026The Art of Efficient Reasoning: Data, Reward, and Optimizationcited by 3
  2. 2026Free(): Learning to Forget in Malloc-Only Reasoning Modelscited by 1
  3. 2026SCA: Segment-Wise CoT Compression with Answer Alignmentcited by 1
  4. 2026Implicit Compression Regularization: Concise Reasoning via Internal Shorter Distributions in RL Post-Trainingcited by 1
  5. 2026Length Penalties Make Chain-of-Thought Less Monitorablecited by 1
  6. 2026Plan Right, Then Plan Tight: Symbolic RL for Efficient Embodied Reasoningcited by 0
  7. 2026QLPO: Quadrant-weighted Sampling for Length-aware Policy Optimizationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026ConMax: Confidence-Maximizing Compression for Efficient Chain-of-Thought Reasoning
  2. 2025Think in Blocks: Adaptive Reasoning from Direct Response to Deep Reasoning
  3. 2025Adaptive Deep Reasoning: Triggering Deep Thinking When Needed
  4. 2025DeepSearch: Overcome the Bottleneck of Reinforcement Learning with Verifiable Rewards via Monte Carlo Tree Search
  5. 2026HMPO: Hybrid Median-length Policy Optimization for Chain-of-Thought Compression
  6. 2025Optimizing Length Compression in Large Reasoning Models
  7. 2025Leash: Adaptive Length Penalty and Reward Shaping for Efficient Large Reasoning Model
  8. 2026Compress the Easy, Explore the Hard: Difficulty-Aware Entropy Regularization for Efficient LLM Reasoning