mlsanityData to 7 Oct 2026

FlashThink: An Early Exit Method For Efficient Reasoning

arXiv 2505.13949 ↗First posted 20 May 2025cs.CL cs.AIcited by 28

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

also in 1 more
Abstract (arXiv)

Large Language Models (LLMs) have shown impressive performance in reasoning tasks. However, LLMs tend to generate excessively long reasoning content, leading to significant computational overhead. Our observations indicate that even on simple problems, LLMs tend to produce unnecessarily lengthy reasoning content, which is against intuitive expectations. Preliminary experiments show that at a certain point during the generation process, the model is already capable of producing the correct solution without completing the full reasoning content. Therefore, we consider that the reasoning process of the model can be exited early to achieve the purpose of efficient reasoning. We introduce a verification model that identifies the exact moment when the model can stop reasoning and still provide the correct answer. Comprehensive experiments on four different benchmarks demonstrate that our proposed method, FlashThink, effectively shortens the reasoning content while preserving the model accuracy. For the Deepseek-R1 and QwQ-32B models, we reduced the length of reasoning content by 77.04% and 77.47%, respectively, without reducing the accuracy.

Key facts. Cited by 28 papers in the map (data through 7 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 29 of its 40 references that are in the map. First posted on arXiv 20 May 2025.

Builds on

29 of 40 references.

Closest ancestors · 11

Same direction, newest first

  1. 2025Reasoning Models Can Be Effective Without Thinking14 Apr
  2. 2025Reasoning Models Know When They're Right: Probing Hidden States for Self-Verification7 Apr
  3. 2025Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models20 Mar
  4. 2025L1: Controlling How Long A Reasoning Model Thinks With Reinforcement Learning6 Mar
  5. 2025DAST: Difficulty-Adaptive Slow-Thinking for Large Reasoning Models6 Mar
  6. 2025LightThinker: Thinking Step-by-Step Compression21 Feb
  7. 2025TokenSkip: Controllable Chain-of-Thought Compression in LLMs17 Feb
  8. 2025Demystifying Long Chain-of-Thought Reasoning in LLMs5 Feb
  9. 2025O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruning22 Jan
  10. 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs30 Dec
  11. 2024Compressed Chain of Thought: Efficient Reasoning Through Dense Representations17 Dec

Other references · 18

  1. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  2. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  3. 2025Qwen3 Technical Reportcited by 8,195
  4. 2021Evaluating Large Language Models Trained on Codecited by 8,053
  5. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  6. 2024Qwen2.5 Technical Reportcited by 4,713
  7. 2023A Survey of Large Language Modelscited by 2,537
  8. 2024OpenAI o1 System Cardcited by 1,220
  9. 2025Kimi k1.5: Scaling Reinforcement Learning with LLMscited by 1,193
  10. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 763
  11. 2025Chain of Draft: Thinking Faster by Writing Lesscited by 290
  12. 2025CODI: Compressing Chain-of-Thought into Continuous Space via Self-Distillationcited by 216
6 more
  1. 2024Teaching Large Language Models to Reason with Reinforcement Learningcited by 158
  2. 2025Phi-4-reasoning Technical Reportcited by 146
  3. 2024Liger Kernel: Efficient Triton Kernels for LLM Trainingcited by 123
  4. 2025Sketch-of-Thought: Efficient LLM Reasoning with Adaptive Cognitive-Inspired Sketchingcited by 116
  5. 2025A Survey of Frontiers in LLM Reasoning: Inference Scaling, Learning to Reason, and Agentic Systemscited by 108
  6. 2024P-ICL: Point In-Context Learning for Named Entity Recognition with Large Language Modelscited by 5

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 12; 2026: 16.

28 papers.

Sort:

By area:

  1. 2025Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computationcited by 120
  2. 2025Don't Overthink It: A Survey of Efficient R1-style Large Reasoning Modelscited by 36
  3. 2025OptimalThinkingBench: Evaluating Over and Underthinking in LLMscited by 35
  4. 2025VCRL: Variance-based Curriculum Reinforcement Learning for Large Language Modelscited by 35
  5. 2025The Price of a Second Thought: On the Evaluation of Reasoning Efficiency in Large Language Modelscited by 17
  6. 2025ARES: Multimodal Adaptive Reasoning via Difficulty-Aware Token-Level Entropy Shapingcited by 16
  7. 2025Retrieval-of-Thought: Efficient Reasoning via Reusing Thoughtscited by 12
  8. 2026Neural Chain-of-Thought Search: Searching the Optimal Reasoning Path to Enhance Large Language Modelscited by 7
  9. 2025On the Provable Performance Guarantee of Efficient Reasoning Modelscited by 6
  10. 2025LYNX: Learning Dynamic Exits for Confidence-Controlled Reasoningcited by 6
  11. 2026ORBIT: On-policy Exploration-Exploitation for Controllable Multi-Budget Reasoningcited by 6
  12. 2026TERMINATOR: Learning Optimal Exit Points for Early Stopping in Chain-of-Thought Reasoningcited by 6
16 more
  1. 2026Mitigating Overthinking in Large Reasoning Language Models via Reasoning Path Deviation Monitoringcited by 6
  2. 2026When Is Thinking Enough? Early Exit via Sufficiency Assessment for Efficient Reasoningcited by 5
  3. 2026Token Economics for LLM Agents: A Dual-View Study from Computing and Economicscited by 5
  4. 2025The Zero-Step Thinking: An Empirical Study of Mode Selection as Harder Early Exit in Reasoning Modelscited by 4
  5. 2026Efficient Reasoning with Balanced Thinkingcited by 4
  6. 2026ESTAR: Early-Stopping Token-Aware Reasoning For Efficient Inferencecited by 3
  7. 2026DVAO: Dynamic Variance-adaptive Advantage Optimization for Multi-reward Reinforcement Learningcited by 3
  8. 2026The Shape of Overthinking: Backtracking Bursts in Long Reasoning Tracescited by 2
  9. 2026Agentic Chain-of-Thought Steering for Efficient and Controllable LLM Reasoningcited by 1
  10. 2025Rethinking Calibration for Early-Exit Neural Networkscited by 0
  11. 2025ThinkBrake: Efficient Reasoning via Log-Probability Margin Guided Decodingcited by 0
  12. 2026FAQ: Mitigating Quantization Error via Regenerating Calibration Data with Family-Aware Quantizationcited by 0
  13. 2026Conditional Performance Guarantee for Large Reasoning Modelscited by 0
  14. 2026Know When To Fold 'Em: Token-Efficient LLM Synthetic Data Generation via Multi-Stage In-Flight Rejectioncited by 0
  15. 2026DyCon: Dynamic Reasoning Control via Evolving Difficulty Modelingcited by 0
  16. 2026OS-Pruner: Pruning Chains-of-Thought of Reasoning Models via Optimal Stoppingcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Dynamic Early Exit in Reasoning Models
  2. 2025Efficient Inference for Large Reasoning Models: A Survey
  3. 2025Advancing Reasoning in Large Language Models: Promising Methods and Approaches
  4. 2025AutoL2S: Auto Long-Short Reasoning for Efficient Large Language Models
  5. 2025Harnessing the Reasoning Economy: A Survey of Efficient Reasoning for Large Language Models
  6. 2026A State-Transition Framework for Efficient LLM Reasoning
  7. 2025A Survey of Efficient Reasoning for Large Reasoning Models: Language, Multimodality, and Beyond
  8. 2026Enhancing reasoning accuracy in large language models during inference time