Controlling Thinking Speed in Reasoning Models

• URL: http://arxiv.org/abs/2507.03704v1
• Date: Fri, 04 Jul 2025 16:41:06 GMT
• Title: Controlling Thinking Speed in Reasoning Models
• Authors: Zhengkai Lin, Zhihang Fu, Ze Chen, Chao Chen, Liang Xie, Wenxiao Wang, Deng Cai, Zheng Wang, Jieping Ye,
• Abstract summary: Human cognition operates in two modes: fast, intuitive System 1 thinking and slow, deliberate System 2 thinking.<n>In this work, we enable LRMs to approximate human intelligence through dynamic thinking speed adjustment.<n>Our approach addresses two key questions: (1) how to control thinking speed in LRMs, and (2) when to adjust it for optimal performance.
• Score: 41.72496532709135
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• 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.

Related papers

• Agentic Test-Time Scaling for WebAgents [65.5178428849495]
  We present Confidence-Aware Test-Time Scaling (CATTS), which uses vote-derived uncertainty to allocate compute only when decisions are genuinely contentious.<n>CATTS improves performance on WebArena-Lite and GoBrowse by up to 9.1% over React while using up to 2.3x fewer tokens than uniform scaling.
  arXiv  Detail & Related papers  (2026-02-12T18:58:30Z)
• Addressing Overthinking in Large Vision-Language Models via Gated Perception-Reasoning Optimization [56.59356959631999]
  Gated Perception-Reasoning Optimization (GPRO) is a meta-reasoning controller that dynamically routes computation among three decision paths.<n>GPRO substantially improves both accuracy and efficiency, outperforming recent slow-thinking methods.
  arXiv  Detail & Related papers  (2026-01-07T23:05:17Z)
• Learning to Think Fast and Slow for Visual Language Models [29.91277432114863]
  We propose a simple RL approach, which enables visual language models to switch between fast and slow thinking modes depending on task difficulty.<n>Our model, named DualMindVLM, significantly outperforms the base model and achieves performance on par with state-of-the-art visual reasoning models.
  arXiv  Detail & Related papers  (2025-11-20T18:59:48Z)
• Adaptive Overclocking: Dynamic Control of Thinking Path Length via Real-Time Reasoning Signals [8.264189366042675]
  We propose Adaptive Overclocking, a method that makes the hyper parameter $alpha$ dynamic and context-aware.<n>Our method adjusts reasoning speed in real time through two complementary signals.<n> Experiments on GSM8K, MATH, and SVAMP show that HAC achieves superior accuracy-latency trade-offs.
  arXiv  Detail & Related papers  (2025-09-21T09:40:27Z)
• Multipole Attention for Efficient Long Context Reasoning [64.94673641704289]
  Large Reasoning Models (LRMs) have shown promising accuracy improvements on complex problem-solving tasks.<n>LRMs need to generate long chain-of-thought reasoning in order to think before answering.<n>We introduce Multipole Attention, which accelerates autoregressive reasoning by only computing exact attention for the most important tokens.
  arXiv  Detail & Related papers  (2025-06-16T03:00:40Z)
• Pangu Embedded: An Efficient Dual-system LLM Reasoner with Metacognition [95.54406667705999]
  Pangu Embedded is an efficient Large Language Model (LLM) reasoner developed on Ascend Neural Processing Units (NPUs)<n>It addresses the significant computational costs and inference latency challenges prevalent in existing reasoning-optimized LLMs.<n>It delivers rapid responses and state-of-the-art reasoning quality within a single, unified model architecture.
  arXiv  Detail & Related papers  (2025-05-28T14:03:02Z)
• Z1: Efficient Test-time Scaling with Code [26.374317704720234]
  Large Language Models (LLMs) can achieve enhanced complex problem-solving through test-time computing scaling.<n>We propose an efficient test-time scaling method that trains LLMs on code-related reasoning trajectories.<n>We present a novel Shifted Thinking Window to mitigate overthinking overhead.
  arXiv  Detail & Related papers  (2025-04-01T14:01:50Z)
• From System 1 to System 2: A Survey of Reasoning Large Language Models [72.87412996793957]
  Foundational Large Language Models excel at fast decision-making but lack depth for complex reasoning.<n>OpenAI's o1/o3 and DeepSeek's R1 have demonstrated expert-level performance in fields such as mathematics and coding.
  arXiv  Detail & Related papers  (2025-02-24T18:50:52Z)
• Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs [76.43407125275202]
  o1-like models can emulate human-like long-time thinking during inference.<n>This paper presents the first comprehensive study on the prevalent issue of overthinking in these models.<n>We propose strategies to mitigate overthinking, streamlining reasoning processes without compromising accuracy.
  arXiv  Detail & Related papers  (2024-12-30T18:55:12Z)
• Efficiently Scaling LLM Reasoning with Certaindex [25.549811985276488]
  Test-time reasoning algorithms can wastefully generate many tokens without improving accuracy.<n>We introduce Certaindex, an algorithm-agnostic metric measuring when further computation is unlikely to alter the final result.<n>Certaindex is lightweight, can accelerate reasoning program inference via early exit, and enables dynamic token allocation.
  arXiv  Detail & Related papers  (2024-12-30T14:57:53Z)
• Understanding Optimization in Deep Learning with Central Flows [53.66160508990508]
  We show that an RMS's implicit behavior can be explicitly captured by a "central flow:" a differential equation. We show that these flows can empirically predict long-term optimization trajectories of generic neural networks.
  arXiv  Detail & Related papers  (2024-10-31T17:58:13Z)
• Tree-of-Mixed-Thought: Combining Fast and Slow Thinking for Multi-hop Visual Reasoning [16.495754104540605]
  Large language models (LLMs) can generate code-like plans for complex inference tasks such as visual reasoning. We propose a hierarchical plan-searching algorithm that integrates the one-stop reasoning (fast) and the Tree-of-thought (slow)
  arXiv  Detail & Related papers  (2023-08-18T16:21:40Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.