mlsanityData to 14 Oct 2026

System-1.5 Reasoning: Traversal in Language and Latent Spaces with Dynamic Shortcuts

arXiv 2505.18962 ↗First posted 25 May 2025cs.CLcited by 27

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

Abstract (arXiv)

Chain-of-thought (CoT) reasoning enables large language models (LLMs) to move beyond fast System-1 responses and engage in deliberative System-2 reasoning. However, this comes at the cost of significant inefficiency due to verbose intermediate output. Recent latent-space reasoning methods improve efficiency by operating on hidden states without decoding into language, yet they treat all steps uniformly, failing to distinguish critical deductions from auxiliary steps and resulting in suboptimal use of computational resources. In this paper, we propose System-1.5 Reasoning, an adaptive reasoning framework that dynamically allocates computation across reasoning steps through shortcut paths in latent space. Specifically, System-1.5 Reasoning introduces two types of dynamic shortcuts. The model depth shortcut (DS) adaptively reasons along the vertical depth by early exiting non-critical tokens through lightweight adapter branches, while allowing critical tokens to continue through deeper Transformer layers. The step shortcut (SS) reuses hidden states across the decoding steps to skip trivial steps and reason horizontally in latent space. Training System-1.5 Reasoning involves a two-stage self-distillation process: first distilling natural language CoT into latent-space continuous thought, and then distilling full-path System-2 latent reasoning into adaptive shortcut paths (System-1.5 Reasoning). Experiments on reasoning tasks demonstrate the superior performance of our method. For example, on GSM8K, System-1.5 Reasoning achieves reasoning performance comparable to traditional CoT fine-tuning methods while accelerating inference by over 20x and reducing token generation by 92.31% on average.

Key facts. Cited by 27 papers in the map (data through 14 Oct 2026). Placed in Prompting and chain-of-thought reasoning, one of 17 directions in LLM reasoning. Builds on 63 of its 78 references that are in the map. First posted on arXiv 25 May 2025.

Builds on

63 of 78 references.

Closest ancestors · 11

Same direction, newest first

  1. 2025Deconstructing Long Chain-of-Thought: A Structured Reasoning Optimization Framework for Long CoT Distillation20 Mar
  2. 2025Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Models12 Mar
  3. 2025CODI: Compressing Chain-of-Thought into Continuous Space via Self-Distillation28 Feb
  4. 2025Chain of Draft: Thinking Faster by Writing Less25 Feb
  5. 2025SoftCoT: Soft Chain-of-Thought for Efficient Reasoning with LLMs17 Feb
  6. 2024Training Large Language Models to Reason in a Continuous Latent Space9 Dec
  7. 2024Distilling System 2 into System 18 Jul
  8. 2024From Explicit CoT to Implicit CoT: Learning to Internalize CoT Step by Step23 May
  9. 2024Let's Think Dot by Dot: Hidden Computation in Transformer Language Models24 Apr
  10. 2023Implicit Chain of Thought Reasoning via Knowledge Distillation2 Nov
  11. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Models28 Jan

Other references · 52

  1. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,305
  2. 2022Training language models to follow instructions with human feedbackcited by 15,971
  3. 2024The Llama 3 Herd of Modelscited by 14,715
  4. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,380
  5. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 13,028
  6. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  7. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,371
  8. 2024GPT-4o System Cardcited by 4,588
  9. 2023SWE-bench: Can Language Models Resolve Real-World GitHub Issues?cited by 3,373
  10. 2023RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Controlcited by 3,080
  11. 2023Voyager: An Open-Ended Embodied Agent with Large Language Modelscited by 2,191
  12. 2023PaLM-E: An Embodied Multimodal Language Modelcited by 2,071
38 more
  1. 2024LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Codecited by 2,004
  2. 2022Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language modelscited by 1,998
  3. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameterscited by 1,986
  4. 2024Mixtral of Expertscited by 1,823
  5. 2023MetaGPT: Meta Programming for A Multi-Agent Collaborative Frameworkcited by 1,808
  6. 2022Fast Inference from Transformers via Speculative Decodingcited by 1,653
  7. 2024Octo: An Open-Source Generalist Robot Policycited by 1,484
  8. 2025s1: Simple test-time scalingcited by 1,341
  9. 2025Kimi k1.5: Scaling Reinforcement Learning with LLMscited by 1,210
  10. 2021Did Aristotle Use a Laptop? A Question Answering Benchmark with Implicit Reasoning Strategiescited by 743
  11. 2025UI-TARS: Pioneering Automated GUI Interaction with Native Agentscited by 610
  12. 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMscited by 570
  13. 2020DeeBERT: Dynamic Early Exiting for Accelerating BERT Inferencecited by 332
  14. 2025From System 1 to System 2: A Survey of Reasoning Large Language Modelscited by 320
  15. 2022Confident Adaptive Language Modelingcited by 310
  16. 2024LayerSkip: Enabling Early Exit Inference and Self-Speculative Decodingcited by 307
  17. 2025O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruningcited by 279
  18. 2023Think before you speak: Training Language Models With Pause Tokenscited by 264
  19. 2024Mixture-of-Depths: Dynamically allocating compute in transformer-based language modelscited by 258
  20. 2019Depth-Adaptive Transformercited by 236
  21. 2024Compressed Chain of Thought: Efficient Reasoning Through Dense Representationscited by 201
  22. 2025Phi-4-reasoning Technical Reportcited by 151
  23. 2023Towards Learning a Generalist Model for Embodied Navigationcited by 138
  24. 2025A Survey of Efficient Reasoning for Large Reasoning Models: Language, Multimodality, and Beyondcited by 132
  25. 2025Advances and Challenges in Foundation Agents: From Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systemscited by 126
  26. 2024Fast Best-of-N Decoding via Speculative Rejectioncited by 112
  27. 2025PRMBench: A Fine-grained and Challenging Benchmark for Process-Level Reward Modelscited by 99
  28. 2025How Well do LLMs Compress Their Own Chain-of-Thought? A Token Complexity Approachcited by 97
  29. 2025Token Assorted: Mixing Latent and Text Tokens for Improved Language Model Reasoningcited by 91
  30. 2023SkipDecode: Autoregressive Skip Decoding with Batching and Caching for Efficient LLM Inferencecited by 89
  31. 2023CoLT5: Faster Long-Range Transformers with Conditional Computationcited by 84
  32. 2024Can Language Models Learn to Skip Steps?cited by 59
  33. 2025Atom of Thoughts for Markov LLM Test-Time Scalingcited by 54
  34. 2024Visual Agents as Fast and Slow Thinkerscited by 53
  35. 2020CascadeBERT: Accelerating Inference of Pre-trained Language Models via Calibrated Complete Models Cascadecited by 47
  36. 2024OSCAR: Operating System Control via State-Aware Reasoning and Re-Planningcited by 38
  37. 2024System-1.x: Learning to Balance Fast and Slow Planning with Language Modelscited by 35
  38. 2025FastMCTS: A Simple Sampling Strategy for Data Synthesiscited by 17

50 most cited of 52 All 52 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 7; 2026: 20.

27 papers.

Sort:

By area:

  1. 2025A Survey on Latent Reasoningcited by 81
  2. 2025Implicit Reasoning in Large Language Models: A Comprehensive Surveycited by 43
  3. 2026The Latent Space: Foundation, Evolution, Mechanism, Ability, and Outlookcited by 43
  4. 2025SwiReasoning: Switch-Thinking in Latent and Explicit for Pareto-Superior Reasoning LLMscited by 22
  5. 2025Towards Concise and Adaptive Thinking in Large Reasoning Models: A Surveycited by 18
  6. 2026Toward Efficient Agents: Memory, Tool learning, and Planningcited by 16
  7. 2025KaVa: Latent Reasoning via Compressed KV-Cache Distillationcited by 8
  8. 2026SeLaR: Selective Latent Reasoning in Large Language Modelscited by 7
  9. 2026Latent Chain-of-Thought as Planning: Decoupling Reasoning from Verbalizationcited by 6
  10. 2025Learning to Ponder: Adaptive Reasoning in Latent Spacecited by 5
  11. 2026SPOT: Span-level Pause-of-Thought for Efficient and Interpretable Latent Reasoning in Large Language Modelscited by 5
  12. 2026LPG: Balancing Efficiency and Policy Reasoning in Latent Policy Guardrailscited by 5
15 more
  1. 2026Self-Policy Distillation via Capability-Selective Subspace Projectioncited by 5
  2. 2026Capabilities and Fundamental Limits of Latent Chain-of-Thoughtcited by 4
  3. 2026LightThinker++: From Reasoning Compression to Memory Managementcited by 4
  4. 2025Optimizing Reasoning Efficiency through Prompt Difficulty Predictioncited by 3
  5. 2026Semantic-Enriched Latent Visual Reasoningcited by 3
  6. 2026Unlocking the Black Box of Latent Reasoning: An Interpretability-Guided Approach to Interventioncited by 3
  7. 2026Latent Abstraction for Retrieval-Augmented Generationcited by 1
  8. 2026Reasoning-aware Speculative Decoding for Efficient Vision-Language-Action Models in Autonomous Drivingcited by 1
  9. 2026Efficient Reasoning on the Edgecited by 0
  10. 2026State of Thought Enables Endogenous Reasoningcited by 0
  11. 2026LatentHarness: Learning Latent Actions for Memory and Reasoning via Counterfactual Policy Distillationcited by 0
  12. 2026MemCodex: Self-Programming Hierarchical Memory for Language Agentscited by 0
  13. 2026SanSi: A Looped Typed Decision Model for System 1.5 Thinkingcited by 0
  14. 2026Coarse-Graining Recursive Expert Computation: Continuation-Relative Interfaces and Failure Modescited by 0
  15. 2026What Makes a Social Agent's Interactions Effective? Learning from Verifiable Outcomes in Paired Worldscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Reasoning Beyond Language: A Comprehensive Survey on Latent Chain-of-Thought Reasoning
  2. 2024Break the Chain: Large Language Models Can be Shortcut Reasoners
  3. 2025Eliciting Chain-of-Thought in Base LLMs via Gradient-Based Representation Optimization
  4. 2025Latent Chain-of-Thought? Decoding the Depth-Recurrent Transformer
  5. 2026How Far Ahead Do LLMs Plan? Uncovering the Latent Horizon in Chain-of-Thought Reasoning
  6. 2026J-CoT: Chain-of-Thought in J-Space
  7. 2024Fine-Tuning on Diverse Reasoning Chains Drives Within-Inference CoT Refinement in LLMs
  8. 2025Can Reasoning Path still be Effective as Input? Bridging Post-Reasoning to Chain-of-Thought Compression