mlsanityData to 14 Oct 2026

Training Large Language Models to Reason in a Continuous Latent Space

arXiv 2412.06769 ↗First posted 9 Dec 2024cs.CLcited by 797

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

Abstract (arXiv)

Large language models (LLMs) are typically constrained to reason in the language space, where they express the reasoning process through a chain-of-thought (CoT) to solve complex problems. However, the language space may not always be optimal for reasoning. Most word tokens primarily ensure textual coherence and are not essential for reasoning, while some critical tokens require complex planning and pose challenges to LLMs. To explore the potential of reasoning beyond language, we introduce a new paradigm called Coconut (Chain of Continuous Thought). Coconut utilizes the last hidden state of the LLM as a representation of the reasoning state, termed "continuous thought." Instead of decoding this state into words, we feed it back to the model as the next input embedding directly in the continuous space. This latent reasoning paradigm enables an advanced reasoning pattern, where continuous thoughts can encode multiple alternative next steps, allowing the model to perform a breadth-first search (BFS) rather than committing prematurely to a single deterministic path as in CoT. Coconut outperforms CoT on logical reasoning tasks that require substantial search during planning and achieves a better trade-off between accuracy and efficiency.

Key facts. Cited by 797 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 43 of its 61 references that are in the map. First posted on arXiv 9 Dec 2024.

Builds on

43 of 61 references.

Closest ancestors · 13

Same direction, newest first

  1. 2024Distilling System 2 into System 18 Jul
  2. 2024From Explicit CoT to Implicit CoT: Learning to Internalize CoT Step by Step23 May
  3. 2024Let's Think Dot by Dot: Hidden Computation in Transformer Language Models24 Apr
  4. 2024Beyond A*: Better Planning with Transformers via Search Dynamics Bootstrapping21 Feb
  5. 2024Chain of Thought Empowers Transformers to Solve Inherently Serial Problems20 Feb
  6. 2023Implicit Chain of Thought Reasoning via Knowledge Distillation2 Nov
  7. 2023Guiding Language Model Reasoning with Planning Tokens9 Oct
  8. 2023Towards Revealing the Mystery behind Chain of Thought: A Theoretical Perspective24 May
  9. 2023Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting7 May
  10. 2022Towards Understanding Chain-of-Thought Prompting: An Empirical Study of What Matters20 Dec
  11. 2022Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango16 Sep
  12. 2022Least-to-Most Prompting Enables Complex Reasoning in Large Language Models21 May
1 more
  1. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Models28 Jan

Other references · 30

  1. 2023GPT-4 Technical Reportcited by 18,396
  2. 2024The Llama 3 Herd of Modelscited by 14,715
  3. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  4. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,379
  5. 2023Tree of Thoughts: Deliberate Problem Solving with Large Language Modelscited by 3,211
  6. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameterscited by 1,986
  7. 2023Math-Shepherd: Verify and Reinforce LLMs Step-by-step without Human Annotationscited by 928
  8. 2023Reasoning with Language Model is Planning with World Modelcited by 861
  9. 2023MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Modelscited by 735
  10. 2019Deep Equilibrium Modelscited by 615
  11. 2022Decomposed Prompting: A Modular Approach for Solving Complex Taskscited by 558
  12. 2023MAmmoTH: Building Math Generalist Models through Hybrid Instruction Tuningcited by 487
18 more
  1. 2024Better & Faster Large Language Models via Multi-token Predictioncited by 360
  2. 2022Language Models Are Greedy Reasoners: A Systematic Formal Analysis of Chain-of-Thoughtcited by 337
  3. 2024Quiet-STaR: Language Models Can Teach Themselves to Think Before Speakingcited by 301
  4. 2023Think before you speak: Training Language Models With Pause Tokenscited by 264
  5. 2023Looped Transformers as Programmable Computerscited by 237
  6. 2022Recurrent Memory Transformercited by 203
  7. 2023Self-Evaluation Guided Beam Search for Reasoningcited by 196
  8. 2023TheoremQA: A Theorem-driven Question Answering datasetcited by 192
  9. 2024Do Large Language Models Latently Perform Multi-Hop Reasoning?cited by 169
  10. 2024Teaching Large Language Models to Reason with Reinforcement Learningcited by 160
  11. 2024Stream of Search (SoS): Learning to Search in Languagecited by 145
  12. 2024Looped Transformers for Length Generalizationcited by 99
  13. 2024Hopping Too Late: Exploring the Limitations of Large Language Models on Multi-Hop Queriescited by 78
  14. 2024LLM Reasoners: New Evaluation, Library, and Analysis of Step-by-Step Reasoning with Large Language Modelscited by 70
  15. 2023Let Models Speak Ciphers: Multiagent Debate through Embeddingscited by 51
  16. 2024Dualformer: Controllable Fast and Slow Thinking by Learning with Randomized Reasoning Tracescited by 43
  17. 2024Distributional reasoning in LLMs: Parallel reasoning processes in multi-hop reasoningcited by 21
  18. 2024Associative Recurrent Memory Transformercited by 15

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 5; 2025: 308; 2026: 484.

797 papers.

Sort:

By area:

  1. 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMscited by 570
  2. 2025Stop Overthinking: A Survey on Efficient Reasoning for Large Language Modelscited by 464
  3. 2025Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approachcited by 415
  4. 2025Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Modelscited by 409
  5. 2025Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Modelscited by 384
  6. 2025From System 1 to System 2: A Survey of Reasoning Large Language Modelscited by 320
  7. 2025Layer by Layer: Uncovering Hidden Representations in Language Modelscited by 311
  8. 2025O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruningcited by 279
  9. 2025Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safetycited by 249
  10. 2024Token-Budget-Aware LLM Reasoningcited by 247
  11. 2025A Survey on Test-Time Scaling in Large Language Models: What, How, Where, and How Well?cited by 223
  12. 2025Dynamic Early Exit in Reasoning Modelscited by 219
38 more
  1. 2025The Landscape of Agentic Reinforcement Learning for LLMs: A Surveycited by 205
  2. 2024Compressed Chain of Thought: Efficient Reasoning Through Dense Representationscited by 201
  3. 2025ThinkPrune: Pruning Long Chain-of-Thought of LLMs via Reinforcement Learningcited by 192
  4. 2025Reasoning Models Can Be Effective Without Thinkingcited by 187
  5. 2025A Survey of Reinforcement Learning for Large Reasoning Modelscited by 180
  6. 2025Hierarchical Reasoning Modelcited by 177
  7. 2025Chain-of-Thought Reasoning In The Wild Is Not Always Faithfulcited by 170
  8. 2025CoT-Valve: Length-Compressible Chain-of-Thought Tuningcited by 169
  9. 2025Scaling Latent Reasoning via Looped Language Modelscited by 165
  10. 2025Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computationcited by 133
  11. 2025Echo Chamber: RL Post-training Amplifies Behaviors Learned in Pretrainingcited by 129
  12. 2025Advances and Challenges in Foundation Agents: From Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systemscited by 126
  13. 2025GRIT: Teaching MLLMs to Think with Imagescited by 125
  14. 2025Machine Mental Imagery: Empower Multimodal Reasoning with Latent Visual Tokenscited by 124
  15. 2025Soft Thinking: Unlocking the Reasoning Potential of LLMs in Continuous Concept Spacecited by 117
  16. 2025SoftCoT: Soft Chain-of-Thought for Efficient Reasoning with LLMscited by 116
  17. 2025A Survey of Frontiers in LLM Reasoning: Inference Scaling, Learning to Reason, and Agentic Systemscited by 111
  18. 2025MemGen: Weaving Generative Latent Memory for Self-Evolving Agentscited by 109
  19. 2025LightThinker: Thinking Step-by-Step Compressioncited by 104
  20. 2025Latent Visual Reasoningcited by 104
  21. 2025GuardReasoner: Towards Reasoning-based LLM Safeguardscited by 100
  22. 2025DeepReview: Improving LLM-based Paper Review with Human-like Deep Thinking Processcited by 100
  23. 2025Efficient Reasoning Models: A Surveycited by 95
  24. 2025DeepSeek-R1 Thoughtology: Let's think about LLM Reasoningcited by 92
  25. 2025Token Assorted: Mixing Latent and Text Tokens for Improved Language Model Reasoningcited by 91
  26. 2025When Chain of Thought is Necessary, Language Models Struggle to Evade Monitorscited by 88
  27. 2025Think Silently, Think Fast: Dynamic Latent Compression of LLM Reasoning Chainscited by 85
  28. 2025Monet: Reasoning in Latent Visual Space Beyond Images and Languagecited by 82
  29. 2025A Survey on Latent Reasoningcited by 81
  30. 2025Reasoning by Superposition: A Theoretical Perspective on Chain of Continuous Thoughtcited by 77
  31. 2025Efficient Reasoning with Hidden Thinkingcited by 75
  32. 2025Thinking with Generated Imagescited by 73
  33. 2025Latent Collaboration in Multi-Agent Systemscited by 73
  34. 2025Dynamic Chunking for End-to-End Hierarchical Sequence Modelingcited by 72
  35. 2025Open Challenges in Multi-Agent Security: Towards Secure Systems of Interacting AI Agentscited by 71
  36. 2025Chain-of-Visual-Thought: Teaching VLMs to See and Think Better with Continuous Visual Tokenscited by 71
  37. 2025SIM-CoT: Supervised Implicit Chain-of-Thoughtcited by 70
  38. 2025SEAL: Steerable Reasoning Calibration of Large Language Models for Freecited by 68

50 most cited of 797 All 797 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2025Reasoning Beyond Language: A Comprehensive Survey on Latent Chain-of-Thought Reasoning
  2. 2025Beyond Chains of Thought: Benchmarking Latent-Space Reasoning Abilities in Large Language Models
  3. 2025Advancing Reasoning in Large Language Models: Promising Methods and Approaches
  4. 2023Chain-of-Thought Hub: A Continuous Effort to Measure Large Language Models' Reasoning Performance
  5. 2025Early Stopping Chain-of-thoughts in Large Language Models
  6. 2025CODI: Compressing Chain-of-Thought into Continuous Space via Self-Distillation
  7. 2025Long Chain-of-Thought Reasoning Across Languages
  8. 2024General Purpose Verification for Chain of Thought Prompting