mlsanityData to 7 Oct 2026

CRISP: Compressed Reasoning via Iterative Self-Policy Distillation

arXiv 2603.05433 ↗First posted 5 Mar 2026cs.LGcited by 49

On the map: LLM reasoning → Prompting and chain-of-thought reasoning

Abstract (arXiv)

Reasoning models often generate far more tokens than a task requires, which raises inference cost and can compound errors. We introduce CRISP (Compressed Reasoning via Iterative Self-Policy Distillation), an on-policy self-distillation method that teaches a model to reason more concisely by distilling its own concise behavior back into itself. The method uses a single idea: condition the same model on a "be concise" instruction to obtain teacher logits, then minimize the per-token reverse KL divergence between the student and this teacher on the student's own rollouts. It requires no ground-truth answers, no token budgets, and no difficulty estimators. The reverse-KL objective is naturally difficulty-adaptive: it compresses easy problems aggressively while preserving the reasoning steps that hard problems require. On Qwen3-14B, CRISP cuts reasoning length by up to 56% on MATH-500 and 38% on the harder AIME 2024, while improving MATH-500 accuracy by up to 3.3 points over the base model and holding AIME 2024 accuracy within about one point. This behavior generalizes across model sizes and families: Qwen3-8B shows the same compression with accuracy preserved, and DeepSeek-R1-Distill-Llama-8B improves accuracy on all five benchmarks while shortening its responses. General capabilities are preserved across all three models. Code is available at https://github.com/HJSang/OPSD_Reasoning_Compression.

Key facts. Cited by 49 papers in the map (data through 7 Oct 2026). Placed in Prompting and chain-of-thought reasoning, one of 17 directions in LLM reasoning. Builds on 35 of its 41 references that are in the map. First posted on arXiv 5 Mar 2026.

Builds on

35 of 41 references.

Closest ancestors · 5

Same direction, newest first

  1. 2026S3-CoT: Self-Sampled Succinct Reasoning Enables Efficient Chain-of-Thought LLMs2 Feb
  2. 2026Self-Distilled Reasoner: On-Policy Self-Distillation for Large Language Models26 Jan
  3. 2025Distilling the Essence: Efficient Reasoning Distillation via Sequence Truncation24 Dec
  4. 2025Concise Reasoning, Big Gains: Pruning Long Reasoning Trace with Difficulty-Aware Prompting26 May
  5. 2025Chain of Draft: Thinking Faster by Writing Less25 Feb

Other references · 30

  1. 2025Qwen3 Technical Reportcited by 8,195
  2. 2020Measuring Massive Multitask Language Understandingcited by 7,768
  3. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,203
  4. 2025Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilitiescited by 3,655
  5. 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,748
  6. 2025DAPO: An Open-Source LLM Reinforcement Learning System at Scalecited by 2,678
  7. 2024HybridFlow: A Flexible and Efficient RLHF Frameworkcited by 1,980
  8. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameterscited by 1,928
  9. 2025s1: Simple test-time scalingcited by 1,311
  10. 2023SGLang: Efficient Execution of Structured Language Model Programscited by 1,107
  11. 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMscited by 564
  12. 2025Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoningcited by 540
18 more
  1. 2025The Entropy Mechanism of Reinforcement Learning for Reasoning Language Modelscited by 463
  2. 2025L1: Controlling How Long A Reasoning Model Thinks With Reinforcement Learningcited by 381
  3. 2026Reinforcement Learning via Self-Distillationcited by 285
  4. 2025TokenSkip: Controllable Chain-of-Thought Compression in LLMscited by 246
  5. 2026Self-Distillation Enables Continual Learningcited by 212
  6. 2025ThinkPrune: Pruning Long Chain-of-Thought of LLMs via Reinforcement Learningcited by 187
  7. 2026On-Policy Context Distillation for Language Modelscited by 161
  8. 2025Wait, We Don't Need to "Wait"! Removing Thinking Tokens Improves Reasoning Efficiencycited by 65
  9. 2026DeepPlanning: Benchmarking Long-Horizon Agentic Planning with Verifiable Constraintscited by 43
  10. 2025The Choice of Divergence: A Neglected Key to Mitigating Diversity Collapse in Reinforcement Learning with Verifiable Rewardcited by 36
  11. 2025DLER: Doing Length pEnalty Right - Incentivizing More Intelligence per Token via Reinforcement Learningcited by 34
  12. 2025TrimR: Verifier-based Training-Free Thinking Compression for Efficient Test-Time Scalingcited by 14
  13. 2025Leash: Adaptive Length Penalty and Reward Shaping for Efficient Large Reasoning Modelcited by 13
  14. 2025The Overthinker's DIET: Cutting Token Calories with DIfficulty-AwarE Trainingcited by 9
  15. 2025Reasoning Efficiently Through Adaptive Chain-of-Thought Compression: A Self-Optimizing Frameworkcited by 6
  16. 2026Overconfident Errors Need Stronger Correction: Asymmetric Confidence Penalties for Reinforcement Learningcited by 6
  17. 2026Mitigating Overthinking in Large Reasoning Models via Difficulty-aware Reinforcement Learningcited by 5
  18. 2026Steering Large Reasoning Models towards Concise Reasoning via Flow Matchingcited by 3

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 49.

49 papers.

Sort:

By area:

  1. 2026Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipecited by 262
  2. 2026TIP: Token Importance in On-Policy Distillationcited by 90
  3. 2026The Many Faces of On-Policy Distillation: Pitfalls, Mechanisms, and Fixescited by 37
  4. 2026Prune-OPD: Efficient and Reliable On-Policy Distillation for Long-Horizon Reasoningcited by 23
  5. 2026Respecting Self-Uncertainty in On-Policy Self-Distillation for Efficient LLM Reasoningcited by 20
  6. 2026Prefix Teach, Suffix Fade: Local Teachability Collapse in Strong-to-Weak On-Policy Distillationcited by 13
  7. 2026ViCuR: Visual Cues as Recoverable Privilege for Multimodal On-Policy Distillationcited by 12
  8. 2026On-Policy Self-Distillation without Any Supervisioncited by 11
  9. 2026Crosslingual On-Policy Self-Distillation for Multilingual Reasoningcited by 10
  10. 2026From Generic Correlation to Input-Specific Credit in On-Policy Self Distillationcited by 9
  11. 2026Beyond GRPO and On-Policy Distillation: An Empirical Sparse-to-Dense Reward Principle for Language-Model Post-Trainingcited by 9
  12. 2026OPRD: On-Policy Representation Distillationcited by 9
37 more
  1. 2026On-Policy Delta Distillationcited by 8
  2. 2026π-Play: Multi-Agent Self-Play via Privileged Self-Distillation without External Datacited by 7
  3. 2026Tailoring Teaching to Aptitude: Direction-Adaptive Self-Distillation for LLM Reasoningcited by 7
  4. 2026OPSD Compresses What RLVR Teaches: A Post-RL Compaction Stage for Reasoning Modelscited by 6
  5. 2026When Should the Teacher Move? Temporal Coupling and Stability in Self On-Policy Distillationcited by 6
  6. 2026DemoPSD: Disagreement-Modulated Policy Self-Distillationcited by 6
  7. 2026Large Language Model Post-Training: A Unified View of Off-Policy and On-Policy Learningcited by 5
  8. 2026Training with Harnesses: On-Policy Harness Self-Distillation for Complex Reasoningcited by 5
  9. 2026A Formula-Driven Survey and Research Agenda for On-Policy Distillationcited by 5
  10. 2026TRIAGE: Role-Typed Credit Assignment for Agentic Reinforcement Learningcited by 5
  11. 2026Answer-Conditioned Chains of Thought Degrade Verifiable-Reasoning Distillation in Large Language Modelscited by 5
  12. 2026Healthcare AI GYM for Medical Agentscited by 4
  13. 2026boldsymbolf-OPD: Stabilizing Long-Horizon On-Policy Distillation with Freshness-Aware Controlcited by 4
  14. 2026Contrastive On-Policy Distillationcited by 3
  15. 2026Reasoning Compression with Mixed-Policy Distillationcited by 2
  16. 2026Restoring the Sweet Spot: Pass-Rate Weighted Self-Distillation for LLM Reasoningcited by 2
  17. 2026Zipping the Thought: When and How Compressed Reasoning Data Works in LLM Post-Trainingcited by 2
  18. 2026Z-Reward: Beyond Scalar Rewards by Internalizing Reasoning into Score Distributionscited by 2
  19. 2026LLM-as-a-Coach: Experiential Learning for Non-Verifiable Taskscited by 2
  20. 2026What Should a Self-Teacher See? Privileged Context Design for On-Policy Self-Distillationcited by 2
  21. 2026Internalize the Temperature: On-Policy Self-Distillation as Policy Reheater for Reinforcement Learningcited by 1
  22. 2026Language Models Need Sleep: Learning to Self-Modify and Consolidate Memoriescited by 1
  23. 2026TurnSight: Turn-Level Hindsight Self-Distillation for Tool-Integrated Reasoningcited by 1
  24. 2026Bidirectional Context Self-Distillation for Reinforcement Learning of Skill-Based LLM Agentscited by 1
  25. 2026Beyond Outcome Rewards: Step-Level Self-Distilled Policy Optimization for Deep Search Agentscited by 1
  26. 2026LatentPress: Context Compression Beyond Text and Visioncited by 1
  27. 2026ClassicLogic: A Knowledge-Driven Benchmark of Classic Puzzle Games for Evaluating Compositional Generalizationcited by 0
  28. 2026OptiAgent: End-to-End Optimization Modeling via Multi-Agent Iterative Refinementcited by 0
  29. 2026Better Starts, Better Ends: Bootstrapped Iterative Self-Reasoning Distillation for Compressed Reasoningcited by 0
  30. 2026Efficient Chain-of-Modality Reasoning via Progressive Compression for Spoken Language Modelscited by 0
  31. 2026RoCo-ACE: Rollout-Conditioned Online Distillation for Retention-Aware Knowledge Injectioncited by 0
  32. 2026SR-OPSD: Self-Referenced On-Policy Self-Distillationcited by 0
  33. 2026iSDFT: Information-Proximal Self-Distillation for Continual Learning in LLMscited by 0
  34. 2026From Self-Distillation to Self-Practice: Privileged Information for Multi-Turn Agentscited by 0
  35. 2026Overcoming Scaling Limits in On-Policy Self-Distillation for LLM Reasoningcited by 0
  36. 2026Mitigating the Length-Scaling Tax with Online Distillationcited by 0
  37. 2026Composing What Each Teacher Learned: Multi-Teacher On-Policy Distillation through Teacher-Relative Shiftscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Learning from a Thoughtful Teacher: Adaptive On-Policy Self-Distillation for Mathematical Reasoning
  2. 2026SimpleOPD: Simple Tokenizer-Agnostic On-Policy Distillation for Long-Context Reasoning
  3. 2026Diagnosing On-Policy Self-Distillation for Reasoning Language Models
  4. 2026Compress-Distill: Reasoning Trace Compression for Efficient Knowledge Distillation
  5. 2026Scaling Reasoning Efficiently via Relaxed On-Policy Distillation
  6. 2026Teach Yourself Where to Look: On-Policy Attention Self-Distillation for Reasoning
  7. 2025ConCISE: Confidence-guided Compression in Step-by-step Efficient Reasoning
  8. 2025Breadcrumbs Reasoning: Memory-Efficient Reasoning with Compression Beacons