A Survey of Efficient Reasoning for Large Reasoning Models: Language, Multimodality, and Beyond

2025/03/27 by Xiaoye Qu, Qu, Xiaoye, Yafu Li +31 · 38 citations
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Natural Language Processing Techniques #Semantic Web and Ontologies #Speech and dialogue systems

paper · pdf · doi:10.48550/arxiv.2503.21614

openalex publication_date 2025/03/27 · openalex created_date 2025/10/11 · openalex updated_date 2026/07/28

Abstract

Recent Large Reasoning Models (LRMs), such as DeepSeek-R1 and OpenAI o1, have demonstrated strong performance gains by scaling up the length of Chain-of-Thought (CoT) reasoning during inference. However, a growing concern lies in their tendency to produce excessively long reasoning traces, which are often filled with redundant content (e.g., repeated definitions), over-analysis of simple problems, and superficial exploration of multiple reasoning paths for harder tasks. This inefficiency introduces significant challenges for training, inference, and real-world deployment (e.g., in agent-based systems), where token economy is critical. In this survey, we provide a comprehensive overview of recent efforts aimed at improving reasoning efficiency in LRMs, with a particular focus on the unique challenges that arise in this new paradigm. We identify common patterns of inefficiency, examine methods proposed across the LRM lifecycle, i.e., from pretraining to inference, and discuss promising future directions for research. To support ongoing development, we also maintain a real-time GitHub repository tracking recent progress in the field. We hope this survey serves as a foundation for further exploration and inspires innovation in this rapidly evolving area.

Cited by

• Offline-Online Curriculum RL for Multimodal Reasoning
• DeepLook: Deeper Thinking with Lookahead
• Understanding Chain-of-Thought in Large Language Models via Topological Data Analysis
• STORM: Search-Guided Generative World Models for Robotic Manipulation
• Anchoring Values in Temporal and Group Dimensions for Flow Matching Model Alignment
• The Vision Wormhole: Latent-Space Communication in Heterogeneous Multi-Agent Systems
• Multimodal Spatial Reasoning in the Large Model Era: A Survey and Benchmarks
• DiffAdapt: Difficulty-Adaptive Reasoning for Token-Efficient LLM Inference
• Higher Satisfaction, Lower Cost: A Technical Report on How LLMs Revolutionize Meituan's Intelligent Interaction Systems
• ExGRPO: Learning to Reason from Experience
• One Token Embedding Is Enough to Deadlock Your Large Reasoning Model
• Plan Then Action:High-Level Planning Guidance Reinforcement Learning for LLM Reasoning
• Spotlight on Token Perception for Multimodal Reinforcement Learning
• Boosting Multi-modal Keyphrase Prediction with Dynamic Chain-of-Thought in Vision-Language Models
• DeepPrune: Parallel Scaling without Inter-trace Redundancy
• ARES: Multimodal Adaptive Reasoning via Difficulty-Aware Token-Level Entropy Shaping
• PEAR: Phase Entropy Aware Reward for Efficient Reasoning
• Native Hybrid Attention for Efficient Sequence Modeling
• Exploring System 1 and 2 communication for latent reasoning in LLMs
• FrameThinker: Learning to Think with Long Videos via Multi-Turn Frame Spotlighting
• Conditional Advantage Estimation for Reinforcement Learning in Large Reasoning Models
• Fast Thinking for Large Language Models
• Your Models Have Thought Enough: Training Large Reasoning Models to Stop Overthinking
• From Long to Lean: Performance-aware and Adaptive Chain-of-Thought Compression via Multi-round Refinement
• Think Smart, Not Hard: Difficulty Adaptive Reasoning for Large Audio Language Models
• DRQA: Dynamic Reasoning Quota Allocation for Controlling Overthinking in Reasoning Large Language Models
• Meta-R1: Empowering Large Reasoning Models with Metacognition
• A Comprehensive Survey on Trustworthiness in Reasoning with Large Language Models
• GLARE: Agentic Reasoning for Legal Judgment Prediction
• Implicit Reasoning in Large Language Models: A Comprehensive Survey
• MIRAGE: Scaling Test-Time Inference with Parallel Graph-Retrieval-Augmented Reasoning Chains
• MME-SCI: A Comprehensive and Challenging Science Benchmark for Multimodal Large Language Models
• RAJ-PGA: Reasoning-Activated Jailbreak and Principle-Guided Alignment Framework for Large Reasoning Models
• Pruning Long Chain-of-Thought of Large Reasoning Models via Small-Scale Preference Optimization
• Pruning the Unsurprising: Efficient LLM Reasoning via First-Token Surprisal
• Efficient Reasoning for Large Reasoning Language Models via Certainty-Guided Reflection Suppression
• Don't Overthink It: A Survey of Efficient R1-style Large Reasoning Models
• C3: A Bilingual Benchmark for Spoken Dialogue Models Exploring Challenges in Complex Conversations

Related

• Efficient Inference for Large Reasoning Models: A Survey
• Don't Overthink It: A Survey of Efficient R1-style Large Reasoning Models
• Towards Large Reasoning Models: A Survey of Reinforced Reasoning with Large Language Models
• Towards Concise and Adaptive Thinking in Large Reasoning Models: A Survey
• Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Models