ThinkRouter: Efficient Reasoning via Routing Thinking between Latent and Discrete Spaces

TL;DR

ThinkRouter is a confidence-aware routing mechanism that dynamically switches between latent and discrete reasoning spaces to improve accuracy and efficiency in large reasoning models.

Contribution

It introduces a novel inference-time routing method that enhances reasoning efficiency by avoiding high-confidence noise and propagating more reliable reasoning trajectories.

Findings

Outperforms explicit CoT and latent reasoning baselines in accuracy.

Achieves an average of 19.70 points improvement in Pass@1.

Reduces generation length by up to 15.55%.

Abstract

Recent work explores latent reasoning to improve reasoning efficiency by replacing explicit reasoning trajectories with continuous representations in a latent space, yet its effectiveness varies across settings. Analysis of model confidence dynamics under latent reasoning reveals that thinking trajectories ending in incorrect answers contain fewer low-confidence steps than those ending in correct answers. Meanwhile, we suggest that soft embeddings aggregated by multiple low-confidence thinking alternatives may introduce and propagate noise, leading to high confidence in unreliable reasoning trajectories. Motivated by these observations, ThinkRouter, an inference-time confidence-aware routing mechanism is proposed to avoid high confidence and noise for efficient reasoning. ThinkRouter routes thinking to the discrete token space when model confidence is low, and to the latent space otherwise. Extensive experiments on STEM reasoning and coding benchmarks across diverse large reasoning models demonstrate that ThinkRouter outperforms explicit CoT, random routing, and latent reasoning baselines in terms of accuracy, achieving an average improvement of 19.70 points in Pass@1, while reducing generation length by up to 15.55%. Further comprehensive analysis reveals that ThinkRouter can calibrate errors arising from explicit CoT and latent reasoning, and accelerates end-of-thinking token generation by globally lowering model confidence.