mlsanityData to 14 Oct 2026

Fast-Slow Thinking GRPO for Large Vision-Language Model Reasoning

arXiv 2504.18458 ↗First posted 25 Apr 2025cs.CL cs.AI cs.CVcited by 36

On the map: LLM reasoning → Reinforcement learning for reasoning

also in 1 more
Abstract (arXiv)

When applying reinforcement learning--typically through GRPO--to large vision-language model reasoning struggles to effectively scale reasoning length or generates verbose outputs across all tasks with only marginal gains in accuracy. To address this issue, we present FAST-GRPO, a variant of GRPO that dynamically adapts reasoning depth based on question characteristics. Through empirical analysis, we establish the feasibility of fast-slow thinking in LVLMs by investigating how response length and data distribution affect performance. Inspired by these observations, we introduce two complementary metrics to estimate the difficulty of the questions, guiding the model to determine when fast or slow thinking is more appropriate. Next, we incorporate adaptive length-based rewards and difficulty-aware KL divergence into the GRPO algorithm. Experiments across seven reasoning benchmarks demonstrate that FAST achieves state-of-the-art accuracy with over 10% relative improvement compared to the base model, while reducing token usage by 32.7-67.3% compared to previous slow-thinking approaches, effectively balancing reasoning length and accuracy.

Key facts. Cited by 36 papers in the map (data through 14 Oct 2026). Placed in Reinforcement learning for reasoning, one of 17 directions in LLM reasoning. Builds on 10 of its 59 references that are in the map. First posted on arXiv 25 Apr 2025.

Builds on

10 of 59 references.

Closest ancestors · 1

Same direction, newest first

  1. 2025Kimi k1.5: Scaling Reinforcement Learning with LLMs22 Jan

Other references · 9

  1. 2025Qwen2.5-VL Technical Reportcited by 5,587
  2. 2024Qwen2 Technical Reportcited by 2,120
  3. 2024HybridFlow: A Flexible and Efficient RLHF Frameworkcited by 2,057
  4. 2025Visual-RFT: Visual Reinforcement Fine-Tuningcited by 513
  5. 2025R1-Onevision: Advancing Generalized Multimodal Reasoning through Cross-Modal Formalizationcited by 370
  6. 2025MM-Eureka: Exploring the Frontiers of Multimodal Reasoning with Rule-based Reinforcement Learningcited by 303
  7. 2025LMM-R1: Empowering 3B LMMs with Strong Reasoning Abilities Through Two-Stage Rule-Based RLcited by 248
  8. 2023Holistic Analysis of Hallucination in GPT-4V(ision): Bias and Interference Challengescited by 131
  9. 2024MMEvalPro: Calibrating Multimodal Benchmarks Towards Trustworthy and Efficient Evaluationcited by 18

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 24; 2026: 12.

36 papers.

Sort:

By area:

  1. 2025More Thinking, Less Seeing? Assessing Amplified Hallucination in Multimodal Reasoning Modelscited by 86
  2. 2025Pangu Embedded: An Efficient Dual-system LLM Reasoner with Metacognitioncited by 62
  3. 2025Reinforced MLLM: A Survey on RL-Based Reasoning in Multimodal Large Language Modelscited by 44
  4. 2025Reasoning on a Budget: A Survey of Adaptive and Controllable Test-Time Compute in LLMscited by 31
  5. 2025MoDoMoDo: Multi-Domain Data Mixtures for Multimodal LLM Reinforcement Learningcited by 30
  6. 2025Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoningcited by 23
  7. 2025A Survey on Agentic Multimodal Large Language Modelscited by 23
  8. 2025Towards Concise and Adaptive Thinking in Large Reasoning Models: A Surveycited by 18
  9. 2025ARES: Multimodal Adaptive Reasoning via Difficulty-Aware Token-Level Entropy Shapingcited by 17
  10. 2025Look Again, Think Slowly: Enhancing Visual Reflection in Vision-Language Modelscited by 16
  11. 2025Fast on the Easy, Deep on the Hard: Efficient Reasoning via Powered Length Penaltycited by 15
  12. 2025Reinforcement Fine-Tuning Powers Reasoning Capability of Multimodal Large Language Modelscited by 14
24 more
  1. 2025SAIL-VL2 Technical Reportcited by 14
  2. 2025Infi-MMR: Curriculum-based Unlocking Multimodal Reasoning via Phased Reinforcement Learning in Multimodal Small Language Modelscited by 10
  3. 2025Understanding and Steering the Cognitive Behaviors of Reasoning Models at Test-Timecited by 10
  4. 2025Interpretable and Reliable Detection of AI-Generated Images via Grounded Reasoning in MLLMscited by 9
  5. 2025Perception-Consistency Multimodal Large Language Models Reasoning via Caption-Regularized Policy Optimizationcited by 9
  6. 2025On the Provable Performance Guarantee of Efficient Reasoning Modelscited by 6
  7. 2026Addressing Overthinking in Large Vision-Language Models via Gated Perception-Reasoning Optimizationcited by 6
  8. 2025Locate-Then-Examine: Grounded Region Reasoning Improves Detection of AI-Generated Imagescited by 4
  9. 2025Enhancing the Outcome Reward-based RL Training of MLLMs with Self-Consistency Samplingcited by 4
  10. 2025FASTopoWM: Fast-Slow Lane Segment Topology Reasoning with Latent World Modelscited by 3
  11. 2025VideoPro: Adaptive Program Reasoning for Long Video Understandingcited by 3
  12. 2025Learning to Think Fast and Slow for Visual Language Modelscited by 3
  13. 2025SSPO: Self-traced Step-wise Preference Optimization for Process Supervision and Reasoning Compressioncited by 2
  14. 2026Beyond Perceptual Shortcuts: Causal-Inspired Debiasing Optimization for Generalizable Video Reasoning in Lightweight MLLMscited by 2
  15. 2026VisionPulse: Dynamic Visual Sparsity for Efficient Multimodal Reasoningcited by 1
  16. 2026Conditional Performance Guarantee for Large Reasoning Modelscited by 0
  17. 2026SesaHand: Enhancing 3D Hand Reconstruction via Controllable Generation with Semantic and Structural Alignmentcited by 0
  18. 2026A Survey of Multimodal Mathematical Reasoning: From Perception, Alignment to Reasoningcited by 0
  19. 2026Thinking Before Matching: A Reinforcement Reasoning Paradigm Towards General Person Re-Identificationcited by 0
  20. 2026SafeLens: Deliberate and Efficient Video Guardrails with Fast-and-Slow Screeningcited by 0
  21. 2026Switch-Reasoner: Learn When to Think in Multitask Mixtures via Reinforcement Learningcited by 0
  22. 2026Stable-MM-R1: Anchoring Multimodal Reasoning Dynamics via Entropy-Guided Stratificationcited by 0
  23. 2026SWEAR: Section-Wise Efficient Audio Reasoning via Self-Rated Difficultycited by 0
  24. 2026Adaptive Look, Flexible Thought: Learning How and How Long to Think in Multimodal Reasoningcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Think or Not? Selective Reasoning via Reinforcement Learning for Vision-Language Models
  2. 2026Transformation-Augmented GRPO for Enhancing Exploration in Reasoning of Large Language Models
  3. 2025VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning
  4. 2025Thinker: Learning to Think Fast and Slow
  5. 2025Predictive Scaling Laws for Efficient GRPO Training of Large Reasoning Models
  6. 2025A Survey of Slow Thinking-based Reasoning LLMs using Reinforced Learning and Inference-time Scaling Law
  7. 2026LEAD: Length-Efficient Adaptive and Dynamic Reasoning for Large Language Models
  8. 2026GRPO Training Dynamics for Small Language Models