Controlling Thinking Speed in Reasoning Models

2025-07-04

AI Summary

Large reasoning models (LRMs) face a fundamental trade-off between rapid response and deep reasoning, resulting in high computational overhead and latency. To address this, we propose a training-free, dynamic thinking-speed control method. Our approach explicitly models “fast/slow thinking” states in representation space via directed guidance vectors and enables online switching of reasoning paths based on real-time task difficulty estimation. Integrated into the vLLM framework, it performs lightweight inference policy adaptation at test time through prompt signals and representation editing—without architectural modification or fine-tuning. Evaluated across mainstream LRMs and multiple benchmarks, our method achieves an average accuracy gain of 1.3% while reducing token consumption by 8.6%, significantly improving the accuracy–efficiency trade-off. Key contributions include: (i) the first explicit representation-space modeling of cognitive speed states via guidance vectors; (ii) real-time, difficulty-aware reasoning-path selection; and (iii) a plug-and-play, zero-shot inference optimization framework compatible with existing LRM deployments.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Search and Optimization: Learning to Search

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• 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

Abstract

Human cognition is theorized to operate in two modes: fast, intuitive System 1 thinking and slow, deliberate System 2 thinking. While current Large Reasoning Models (LRMs) excel at System 2 thinking, their inability to perform fast thinking leads to high computational overhead and latency. In this work, we enable LRMs to approximate human intelligence through dynamic thinking speed adjustment, optimizing accuracy-efficiency trade-offs. Our approach addresses two key questions: (1) how to control thinking speed in LRMs, and (2) when to adjust it for optimal performance. For the first question, we identify the steering vector that governs slow-fast thinking transitions in LRMs' representation space. Using this vector, we achieve the first representation editing-based test-time scaling effect, outperforming existing prompt-based scaling methods. For the second question, we apply real-time difficulty estimation to signal reasoning segments of varying complexity. Combining these techniques, we propose the first reasoning strategy that enables fast processing of easy steps and deeper analysis for complex reasoning. Without any training or additional cost, our plug-and-play method yields an average +1.3% accuracy with -8.6% token usage across leading LRMs and advanced reasoning benchmarks. All of our algorithms are implemented based on vLLM and are expected to support broader applications and inspire future research.

Problem

• Enable LRMs to adjust thinking speed dynamically
• Control slow-fast thinking transitions in LRMs
• Optimize accuracy-efficiency trade-offs in reasoning

Innovation

• Dynamic thinking speed adjustment in LRMs
• Representation editing-based test-time scaling
• Real-time difficulty estimation for reasoning