A Survey on Latent Reasoning

2025-07-08

AI Summary

Explicit chain-of-thought (CoT) reasoning is constrained by the expressive bandwidth of natural language and its reliance on token-level supervision. Method: This work systematically investigates implicit reasoning—multi-step, language-free inference conducted directly within the continuous hidden states of large language models (LLMs). We propose a unified analytical framework and introduce an infinite-depth implicit reasoning mechanism based on masked diffusion models, enabling globally consistent, invertible, and supervision-free inference. Furthermore, we model hierarchical reasoning at the neural level via activation recurrence, latent-state propagation, and trajectory-internalized fine-tuning. Contribution/Results: Our work establishes the first comprehensive survey and knowledge framework for implicit reasoning techniques and releases an open-source GitHub repository to advance the field.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Large Language Models (LLMs) have demonstrated impressive reasoning capabilities, especially when guided by explicit chain-of-thought (CoT) reasoning that verbalizes intermediate steps. While CoT improves both interpretability and accuracy, its dependence on natural language reasoning limits the model's expressive bandwidth. Latent reasoning tackles this bottleneck by performing multi-step inference entirely in the model's continuous hidden state, eliminating token-level supervision. To advance latent reasoning research, this survey provides a comprehensive overview of the emerging field of latent reasoning. We begin by examining the foundational role of neural network layers as the computational substrate for reasoning, highlighting how hierarchical representations support complex transformations. Next, we explore diverse latent reasoning methodologies, including activation-based recurrence, hidden state propagation, and fine-tuning strategies that compress or internalize explicit reasoning traces. Finally, we discuss advanced paradigms such as infinite-depth latent reasoning via masked diffusion models, which enable globally consistent and reversible reasoning processes. By unifying these perspectives, we aim to clarify the conceptual landscape of latent reasoning and chart future directions for research at the frontier of LLM cognition. An associated GitHub repository collecting the latest papers and repos is available at: https://github.com/multimodal-art-projection/LatentCoT-Horizon/.

Problem

• Latent reasoning overcomes CoT's language bandwidth limits
• Exploring neural layers as reasoning computational substrate
• Surveying advanced latent reasoning methods for LLMs

Innovation

• Latent reasoning uses hidden state for inference
• Hierarchical representations enable complex transformations
• Infinite-depth reasoning via masked diffusion models