ThinkFuse: Trajectory-Aware Test-Time Fusion for Small Reasoning Models

2026-10-06

AI Summary

This study addresses the vulnerability of small reasoning models to erroneous reasoning trajectories and the susceptibility of existing test-time fusion methods to misleading local, transient uncertainty. To overcome these limitations, this work proposes a training-free, trajectory-aware test-time fusion framework that leverages trajectory-level uncertainty trends as a global reference. By comparing segment-level and trajectory-level uncertainty discrepancies, the method precisely identifies unstable reasoning points and selectively injects auxiliary paths for correction. This approach effectively mitigates interference from local signals while significantly reducing fusion trigger frequency and generated token count. Comprehensive evaluations on mathematical and knowledge-intensive reasoning benchmarks demonstrate consistent superiority over baselines, yielding robust gains across diverse model combinations. Notably, the framework exhibits particularly efficient and stable performance when applied with smaller primary models.

Abstract

Small reasoning models (SRMs) have shown strong performance on complex reasoning tasks by generating extended chain-of-thought trajectories, but they often fail to recover once their reasoning enters an erroneous path. Existing test-time fusion methods rely on local fusion signals to determine when to trigger fusion, which can be misled by transient uncertainty fluctuations and may reinforce unstable reasoning trajectories. We propose ThinkFuse, a training-free test-time fusion framework that selectively intervenes in unreliable reasoning segments. ThinkFuse compares segment-level uncertainty shifts with trajectory-level uncertainty trends to identify unstable reasoning points and fuse auxiliary reasoning paths into the primary model's trajectory. Extensive experiments demonstrate that ThinkFuse outperforms baselines on mathematical and knowledge-intensive reasoning benchmarks, with consistent gains across model-family combinations, and remains robust with a smaller primary model. Our analysis shows that ThinkFuse requires fewer fusion triggers and generates fewer tokens, highlighting the efficiency of selective triggering. Our code is available at https://github.com/js-lee-AI/ThinkFuse.

Problem

• Small Reasoning Models
• Test-Time Fusion
• Chain-of-Thought
• Reasoning Trajectory
• Uncertainty

Innovation

• Test-Time Fusion
• Small Reasoning Models
• Trajectory-Aware
• Uncertainty Estimation
• Training-Free