SAT: Balancing Reasoning Accuracy and Efficiency with Stepwise Adaptive Thinking

TL;DR

SAT is a framework that adaptively prunes reasoning steps in large models, reducing token usage by up to 40% while maintaining or improving accuracy across multiple benchmarks.

Contribution

It introduces a stepwise, difficulty-aware pruning method using a finite-state machine and a process reward model to optimize reasoning efficiency.

Findings

Achieves up to 40% reduction in reasoning tokens.

Maintains or improves accuracy across 9 models and 7 benchmarks.

Demonstrates effectiveness of adaptive, difficulty-aware reasoning pruning.

Abstract

Large Reasoning Models (LRMs) have revolutionized complex problem-solving, yet they exhibit a pervasive "overthinking", generating unnecessarily long reasoning chains. While current solutions improve token efficiency, they often sacrifice fine-grained control or risk disrupting the logical integrity of the reasoning process. To address this, we introduce Stepwise Adaptive Thinking (SAT), a framework that performs step-level, difficulty-aware pruning while preserving the core reasoning structure. SAT formulates reasoning as a Finite-State Machine (FSM) with distinct thinking modes (Slow, Normal, Fast, Skip). It navigates these states dynamically using a lightweight Process Reward Model (PRM), compressing easy steps while preserving depth for hard ones. Experiments across 9 LRMs and 7 benchmarks show that SAT achieves up to 40% reduction in reasoning tokens while generally maintaining or improving accuracy.