mlsanityData to 7 Oct 2026

Capabilities and Fundamental Limits of Latent Chain-of-Thought

arXiv 2602.01148 ↗First posted 1 Feb 2026cs.AI cs.IT cs.LGcited by 3

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

Abstract (arXiv)

Latent Chain-of-Thought (Latent CoT) models promise efficient reasoning via continuous representations, yet exhibit puzzling performance inconsistencies: excelling at exploration (ProsQA: 97.0%) but failing at computation (GSM8K: 34.1%). We reveal that this trade-off is governed by decisional certainty. Our contributions are threefold: (1) We theoretically characterize the fundamental Exploration-Execution Trade-off, proving that high certainty enables precise execution but inhibits exploration, while low certainty facilitates search but causes error accumulation. (2) We introduce the Symbolic Index--quantifying decisional commitment--as the core mechanism governing this trade-off and establish its causal relationship with both execution stability and exploration capability. (3) We prove that curriculum learning is theoretically necessary, as direct training provably fails due to distributional mismatch. Our framework shifts the design paradigm from binary architectural choices toward adaptive systems that dynamically regulate decisional certainty based on task demands.

Key facts. Cited by 3 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 39 of its 50 references that are in the map. First posted on arXiv 1 Feb 2026.

Builds on

39 of 50 references.

Closest ancestors · 14

Same direction, newest first

  1. 2025SynAdapt: Learning Adaptive Reasoning in Large Language Models via Synthetic Continuous Chain-of-Thought1 Aug
  2. 2025When Chain of Thought is Necessary, Language Models Struggle to Evade Monitors7 Jul
  3. 2025Continuous Chain of Thought Enables Parallel Exploration and Reasoning29 May
  4. 2025System-1.5 Reasoning: Traversal in Language and Latent Spaces with Dynamic Shortcuts25 May
  5. 2025Soft Thinking: Unlocking the Reasoning Potential of LLMs in Continuous Concept Space21 May
  6. 2025Reasoning by Superposition: A Theoretical Perspective on Chain of Continuous Thought18 May
  7. 2025CODI: Compressing Chain-of-Thought into Continuous Space via Self-Distillation28 Feb
  8. 2025SoftCoT: Soft Chain-of-Thought for Efficient Reasoning with LLMs17 Feb
  9. 2025Enhancing Auto-regressive Chain-of-Thought through Loop-Aligned Reasoning12 Feb
  10. 2024Training Large Language Models to Reason in a Continuous Latent Space9 Dec
  11. 2024Expediting and Elevating Large Language Model Reasoning via Hidden Chain-of-Thought Decoding13 Sep
  12. 2024From Explicit CoT to Implicit CoT: Learning to Internalize CoT Step by Step23 May
2 more
  1. 2023Implicit Chain of Thought Reasoning via Knowledge Distillation2 Nov
  2. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Models21 Mar

Other references · 25

  1. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  2. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  3. 2015Deep Learning and the Information Bottleneck Principlecited by 793
  4. 2024LLaVA-CoT: Let Vision Language Models Reason Step-by-Stepcited by 446
  5. 2024Compressed Chain of Thought: Efficient Reasoning Through Dense Representationscited by 192
  6. 2025CoT-Valve: Length-Compressible Chain-of-Thought Tuningcited by 167
  7. 2025LightThinker: Thinking Step-by-Step Compressioncited by 104
  8. 2025Token Assorted: Mixing Latent and Text Tokens for Improved Language Model Reasoningcited by 87
  9. 2025Think Silently, Think Fast: Dynamic Latent Compression of LLM Reasoning Chainscited by 81
  10. 2025Efficient Reasoning with Hidden Thinkingcited by 72
  11. 2023A Survey of Reasoning with Foundation Modelscited by 70
  12. 2025A Survey of Personalized Large Language Models: Progress and Future Directionscited by 65
13 more
  1. 2024Language Models are Hidden Reasoners: Unlocking Latent Reasoning Capabilities via Self-Rewardingcited by 53
  2. 2023Towards a Mechanistic Interpretation of Multi-Step Reasoning Capabilities of Language Modelscited by 50
  3. 2025PonderLM: Pretraining Language Models to Ponder in Continuous Spacecited by 44
  4. 2025Reasoning to Learn from Latent Thoughtscited by 43
  5. 2025SoftCoT++: Test-Time Scaling with Soft Chain-of-Thought Reasoningcited by 38
  6. 2025Don't Overthink It: A Survey of Efficient R1-style Large Reasoning Modelscited by 36
  7. 2025LLM Pretraining with Continuous Conceptscited by 33
  8. 2025Latent Thought Models with Variational Bayes Inference-Time Computationcited by 25
  9. 2024Thinking Tokens for Language Modelingcited by 23
  10. 2025How do Transformers Learn Implicit Reasoning?cited by 15
  11. 2025Enhancing Latent Computation in Transformers with Latent Tokenscited by 13
  12. 2025Reconsidering Overthinking: Penalizing Internal and External Redundancy in CoT Reasoningcited by 13
  13. 2025Beyond Words: A Latent Memory Approach to Internal Reasoning in LLMscited by 7

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 3.

3 papers.

  1. 2026Bridging the Gap Between Latent and Explicit Reasoning with Looped Transformerscited by 11
  2. 2026Training Continuous Chain of Thought Models: A Tale of Two Regimescited by 2
  3. 2026Why Struggle with Continuous Latents? Interpretable Discrete Latent Reasoning via Rendered Compressioncited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025CTRLS: Chain-of-Thought Reasoning via Latent State-Transition
  2. 2026Free Energy Heuristics: Fast-And-Frugal Cognition as Active Inference Under Uncertain Precision
  3. 2026Emergent Search and Backtracking in Latent Reasoning Models
  4. 2026Latent Chain-of-Thought as Planning: Decoupling Reasoning from Verbalization
  5. 2026How Far Ahead Do LLMs Plan? Uncovering the Latent Horizon in Chain-of-Thought Reasoning
  6. 2026Do Latent-CoT Models Think Step-by-Step? A Mechanistic Study on Sequential Reasoning Tasks
  7. 2026Reasoning aligns language models to human cognition
  8. 2025A Comment On "The Illusion of Thinking": Reframing the Reasoning Cliff as an Agentic Gap