Latent Chain-of-Thought as Planning: Decoupling Reasoning from Verbalization

TL;DR

PLaT reframes reasoning as latent planning by decoupling thinking from language. A deterministic Planner evolves a trajectory of continuous latent states $\{\mathbf{S}_k\}$, while a separate Decoder grounds these plans into text, enabling dynamic termination without fixed latent steps. Training combines reconstruction-based supervision with Gaussian denoising, and reinforcement learning via GRPO refines the decoding policy while keeping the planner fixed. Empirically, PLaT achieves superior Pass@$k$ scalability and broader solution diversity on mathematical benchmarks, offering a transparent, scalable foundation for inference-time search despite a modest drop in greedy accuracy compared to explicit CoT.

Abstract

Chain-of-Thought (CoT) empowers Large Language Models (LLMs) to tackle complex problems, but remains constrained by the computational cost and reasoning path collapse when grounded in discrete token spaces. Recent latent reasoning approaches attempt to optimize efficiency by performing reasoning within continuous hidden states. However, these methods typically operate as opaque end-to-end mappings from explicit reasoning steps to latent states, and often require a pre-defined number of latent steps during inference. In this work, we introduce PLaT (Planning with Latent Thoughts), a framework that reformulates latent reasoning as planning by fundamentally decouple reasoning from verbalization. We model reasoning as a deterministic trajectory of latent planning states, while a separate Decoder grounds these thoughts into text when necessary. This decoupling allows the model to dynamically determine when to terminate reasoning rather than relying on fixed hyperparameters. Empirical results on mathematical benchmarks reveal a distinct trade-off: while PLaT achieves lower greedy accuracy than baselines, it demonstrates superior scalability in terms of reasoning diversity. This indicates that PLaT learns a robust, broader solution space, offering a transparent and scalable foundation for inference-time search.

Figures (13)

• Figure 1: Comparison of PLaT and other reasoning strategies. CoT is an explicit chain-of-thought reasoning method, and the rest are implicit latent reasoning methods.
• Figure 1: Confusion matrix of Human-LLM agreement.
• Figure 2: Framework of the proposed PLaT paradigm. (1) SFT Stage: the Planner autoregressively steps forward to generate the latent states in the context of the question. The Decoder then utilizes the projected latent states as the prefix to verbalize them. (2) RL Stage: the Decoder decodes the same states with a sampling strategy to roll out different results. Equations that are valid in the corresponding reasoning process and correct answers reinforce the Decoder as a policy.
• Figure 2: Evolution of Token Distribution Entropy over Normalized Reasoning Progress. The X-axis represents the relative progress of the reasoning chain generation ($0 \to 100\%$), and the Y-axis represents the entropy of the Decoder's output distribution.
• Figure 3: Scaling properties of reasoning diversity across datasets. PLaT-1 and PLaT-2 are the results of PLaT when $N_L=1$ and $N_L=2$, respectively.