mlsanityData to 14 Oct 2026

Prolonged Reasoning Is Not All You Need: Certainty-Based Adaptive Routing for Efficient LLM/MLLM Reasoning

arXiv 2505.15154 ↗First posted 21 May 2025cs.CL cs.AI cs.MMcited by 35

On the map: LLM reasoning → Test-time compute and efficient thinking

also in 1 more
Abstract (arXiv)

Recent advancements in reasoning have significantly enhanced the capabilities of Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) across diverse tasks. However, excessive reliance on chain-of-thought (CoT) reasoning can impair model performance and brings unnecessarily lengthened outputs, reducing efficiency. Our work reveals that prolonged reasoning does not universally improve accuracy and even degrade performance on simpler tasks. To address this, we propose Certainty-based Adaptive Reasoning (CAR), a novel framework that dynamically switches between short answers and long-form reasoning based on the model perplexity. CAR first generates a short answer and evaluates its perplexity, triggering reasoning only when the model exhibits low confidence (i.e., high perplexity). Experiments across diverse multimodal VQA/KIE benchmarks and text reasoning datasets show that CAR outperforms both short-answer and long-form reasoning approaches, striking an optimal balance between accuracy and efficiency.

Key facts. Cited by 35 papers in the map (data through 14 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 30 of its 69 references that are in the map. First posted on arXiv 21 May 2025.

Builds on

30 of 69 references.

Closest ancestors · 7

Same direction, newest first

  1. 2025Efficient Reasoning Models: A Survey15 Apr
  2. 2025Reasoning Models Can Be Effective Without Thinking14 Apr
  3. 2025Reasoning to Learn from Latent Thoughts24 Mar
  4. 2025Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models20 Mar
  5. 2025How Well do LLMs Compress Their Own Chain-of-Thought? A Token Complexity Approach3 Mar
  6. 2024Token-Budget-Aware LLM Reasoning24 Dec
  7. 2024Concise Thoughts: Impact of Output Length on LLM Reasoning and Cost29 Jul

Other references · 23

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 13,028
  3. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  4. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,379
  5. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  6. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,708
  7. 2016SGDR: Stochastic Gradient Descent with Warm Restartscited by 5,045
  8. 2024Qwen2.5 Technical Reportcited by 4,942
  9. 2024Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolutioncited by 4,234
  10. 2017Decoupled Weight Decay Regularizationcited by 1,831
  11. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 797
  12. 2024DeepSeek-VL: Towards Real-World Vision-Language Understandingcited by 733
11 more
  1. 2025VLM-R1: A Stable and Generalizable R1-style Large Vision-Language Modelcited by 526
  2. 2025Chain of Draft: Thinking Faster by Writing Lesscited by 297
  3. 2025CODI: Compressing Chain-of-Thought into Continuous Space via Self-Distillationcited by 225
  4. 2025Seed1.5-Thinking: Advancing Superb Reasoning Models with Reinforcement Learningcited by 161
  5. 2023DocPedia: Unleashing the Power of Large Multimodal Model in the Frequency Domain for Versatile Document Understandingcited by 106
  6. 2023UniDoc: A Universal Large Multimodal Model for Simultaneous Text Detection, Recognition, Spotting and Understandingcited by 74
  7. 2024A Bounding Box is Worth One Token: Interleaving Layout and Text in a Large Language Model for Document Understandingcited by 57
  8. 2025Vision as LoRAcited by 37
  9. 2025Advancing Sequential Numerical Prediction in Autoregressive Modelscited by 26
  10. 2024Boosting of Thoughts: Trial-and-Error Problem Solving with Large Language Modelscited by 22
  11. 2025EVE: Towards End-to-End Video Subtitle Extraction with Vision-Language Modelscited by 6

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 9; 2026: 26.

35 papers.

Sort:

By area:

  1. 2025Don't Overthink It: A Survey of Efficient R1-style Large Reasoning Modelscited by 37
  2. 2025Towards Concise and Adaptive Thinking in Large Reasoning Models: A Surveycited by 18
  3. 2025MEML-GRPO: Heterogeneous Multi-Expert Mutual Learning for RLVR Advancementcited by 14
  4. 2025Aware First, Think Less: Dynamic Boundary Self-Awareness Drives Extreme Reasoning Efficiency in Large Language Modelscited by 13
  5. 2025SABER: Switchable and Balanced Training for Efficient LLM Reasoningcited by 12
  6. 2026Efficient Inference for Large Vision-Language Models: Bottlenecks, Techniques, and Prospectscited by 9
  7. 2026Neural Chain-of-Thought Search: Searching the Optimal Reasoning Path to Enhance Large Language Modelscited by 8
  8. 2026iReasoner: Trajectory-Aware Intrinsic Reasoning Supervision for Self-Evolving Large Multimodal Modelscited by 7
  9. 2026Think-as-You-See: Streaming Chain-of-Thought Reasoning for Large Vision-Language Modelscited by 6
  10. 2025Verifying Meta-Awareness via Predictive Rewards in Reasoning Modelscited by 3
  11. 2025Learning to Reason Efficiently with Discounted Reinforcement Learningcited by 3
  12. 2025Can Confidence Estimates Decide When Chain-of-Thought Is Necessary for LLMs?cited by 2
23 more
  1. 2026Don't Blink: Evidence Collapse during Multimodal Reasoningcited by 2
  2. 2025From Perception to Reasoning: Deep Thinking Empowers Multimodal Large Language Modelscited by 1
  3. 2026LPCAN: Lightweight Pyramid Cross-Attention Network for Rail Surface Defect Detection Using RGB-D Datacited by 0
  4. 2026Knowledge-Embedded and Hypernetwork-Guided Few-Shot Substation Meter Defect Image Generation Methodcited by 0
  5. 2026MA-LipNet: Multi-Dimensional Attention Networks for Robust Lipreadingcited by 0
  6. 2026HyPER: Bridging Exploration and Exploitation for Scalable LLM Reasoning with Hypothesis Path Expansion and Reductioncited by 0
  7. 2026Bridging Efficiency and Transparency: Explainable CoT Compression in Multimodal Large Reasoning Modelscited by 0
  8. 2026The Diffusion Duet: Harmonizing Dual Channels with Wavelet Suppression for Image Separationcited by 0
  9. 2026Detecting Deepfakes with Multivariate Soft Blending and CLIP-based Image-Text Alignmentcited by 0
  10. 2026Efficient Image Super-Resolution with Multi-Scale Spatial Adaptive Attention Networkscited by 0
  11. 2026A Deformable Attention-Based Detection Transformer with Cross-Scale Feature Fusion for Industrial Coil Spring Inspectioncited by 0
  12. 2026Transmittance-Guided Structure-Texture Decomposition for Nighttime Image Dehazingcited by 0
  13. 2026Hierarchical Awareness Adapters with Hybrid Pyramid Feature Fusion for Dense Depth Predictioncited by 0
  14. 2026Breaking the Reward Barrier: Accelerating Tree-of-Thought Reasoning via Speculative Explorationcited by 0
  15. 2026Local Spatiotemporal Convolutional Network for Robust Gait Recognitioncited by 0
  16. 2026Do You Need Text Rectification? Soft Attention Mask Embedding for Rectification-Free Scene Text Spottingcited by 0
  17. 2026DAIN: Dynamic Agent-Based Interaction Network for Efficient and Collaborative Multimodal Reasoningcited by 0
  18. 2026ProWAFT: A ROMA-LPD Instance for Workload-Aware and Dynamic Fault Tolerance in FPGA-Based CNN Acceleratorscited by 0
  19. 2026Offline-Online Curriculum RL for Multimodal Reasoningcited by 0
  20. 2026Rank-Consistent Set Reasoning for Co-Salient Object Detectioncited by 0
  21. 2026Collapse, Not Complexity: Failure-Conditioned Decomposition Repair for End-to-End Document Parsingcited by 0
  22. 2026When Should Agents Think? Adaptive Reasoning via Cross-Turn Estimationcited by 0
  23. 2026SWEAR: Section-Wise Efficient Audio Reasoning via Self-Rated Difficultycited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Certainty-Guided Reasoning in Large Language Models: A Dynamic Thinking Budget Approach
  2. 2026Correct, Concise and Complete: Multi-stage Training For Adaptive Reasoning
  3. 2025Route to Reason: Adaptive Routing for LLM and Reasoning Strategy Selection
  4. 2026Enhancing reasoning accuracy in large language models during inference time
  5. 2025AutoL2S: Auto Long-Short Reasoning for Efficient Large Language Models
  6. 2026ODAR: Principled Adaptive Routing for LLM Reasoning via Active Inference
  7. 2026Stabilizing Efficient Reasoning with Step-Level Advantage Selection
  8. 2026Improving reasoning at inference time via uncertainty minimisation