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

arXiv: https://arxiv.org/abs/2504.18458 · HTML: https://mlsanity.com/paper/2504.18458/

## 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.

## Placed in

- LLM reasoning / [Reinforcement learning for reasoning](https://mlsanity.com/lr/reinforcement-learning-for-reasoning/)
- Multimodal LLMs / vision-language models / [Embodied and 3D vision-language models](https://mlsanity.com/mllm/embodied-and-3d-vision-language-models/)

## Builds on (most cited first, 10 of 10)

- [Qwen2.5-VL Technical Report](https://mlsanity.com/paper/2502.13923/) (2025)
- [Qwen2 Technical Report](https://mlsanity.com/paper/2407.10671/) (2024)
- [HybridFlow: A Flexible and Efficient RLHF Framework](https://mlsanity.com/paper/2409.19256/) (2024)
- [Kimi k1.5: Scaling Reinforcement Learning with LLMs](https://mlsanity.com/paper/2501.12599/) (2025)
- [Visual-RFT: Visual Reinforcement Fine-Tuning](https://mlsanity.com/paper/2503.01785/) (2025)
- [R1-Onevision: Advancing Generalized Multimodal Reasoning through Cross-Modal Formalization](https://mlsanity.com/paper/2503.10615/) (2025)
- [MM-Eureka: Exploring the Frontiers of Multimodal Reasoning with Rule-based Reinforcement Learning](https://mlsanity.com/paper/2503.07365/) (2025)
- [LMM-R1: Empowering 3B LMMs with Strong Reasoning Abilities Through Two-Stage Rule-Based RL](https://mlsanity.com/paper/2503.07536/) (2025)
- [Holistic Analysis of Hallucination in GPT-4V(ision): Bias and Interference Challenges](https://mlsanity.com/paper/2311.03287/) (2023)
- [MMEvalPro: Calibrating Multimodal Benchmarks Towards Trustworthy and Efficient Evaluation](https://mlsanity.com/paper/2407.00468/) (2024)

## Built on by (most cited first, 30 of 36)

- [More Thinking, Less Seeing? Assessing Amplified Hallucination in Multimodal Reasoning Models](https://mlsanity.com/paper/2505.21523/) (2025)
- [Pangu Embedded: An Efficient Dual-system LLM Reasoner with Metacognition](https://mlsanity.com/paper/2505.22375/) (2025)
- [Reinforced MLLM: A Survey on RL-Based Reasoning in Multimodal Large Language Models](https://mlsanity.com/paper/2504.21277/) (2025)
- [Reasoning on a Budget: A Survey of Adaptive and Controllable Test-Time Compute in LLMs](https://mlsanity.com/paper/2507.02076/) (2025)
- [MoDoMoDo: Multi-Domain Data Mixtures for Multimodal LLM Reinforcement Learning](https://mlsanity.com/paper/2505.24871/) (2025)
- [Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoning](https://mlsanity.com/paper/2508.21048/) (2025)
- [A Survey on Agentic Multimodal Large Language Models](https://mlsanity.com/paper/2510.10991/) (2025)
- [Towards Concise and Adaptive Thinking in Large Reasoning Models: A Survey](https://mlsanity.com/paper/2507.09662/) (2025)
- [ARES: Multimodal Adaptive Reasoning via Difficulty-Aware Token-Level Entropy Shaping](https://mlsanity.com/paper/2510.08457/) (2025)
- [Look Again, Think Slowly: Enhancing Visual Reflection in Vision-Language Models](https://mlsanity.com/paper/2509.12132/) (2025)
- [Fast on the Easy, Deep on the Hard: Efficient Reasoning via Powered Length Penalty](https://mlsanity.com/paper/2506.10446/) (2025)
- [Reinforcement Fine-Tuning Powers Reasoning Capability of Multimodal Large Language Models](https://mlsanity.com/paper/2505.18536/) (2025)
- [SAIL-VL2 Technical Report](https://mlsanity.com/paper/2509.14033/) (2025)
- [Infi-MMR: Curriculum-based Unlocking Multimodal Reasoning via Phased Reinforcement Learning in Multimodal Small Language Models](https://mlsanity.com/paper/2505.23091/) (2025)
- [Understanding and Steering the Cognitive Behaviors of Reasoning Models at Test-Time](https://mlsanity.com/paper/2512.24574/) (2025)
- [Interpretable and Reliable Detection of AI-Generated Images via Grounded Reasoning in MLLMs](https://mlsanity.com/paper/2506.07045/) (2025)
- [Perception-Consistency Multimodal Large Language Models Reasoning via Caption-Regularized Policy Optimization](https://mlsanity.com/paper/2509.21854/) (2025)
- [On the Provable Performance Guarantee of Efficient Reasoning Models](https://mlsanity.com/paper/2510.09133/) (2025)
- [Addressing Overthinking in Large Vision-Language Models via Gated Perception-Reasoning Optimization](https://mlsanity.com/paper/2601.04442/) (2026)
- [Locate-Then-Examine: Grounded Region Reasoning Improves Detection of AI-Generated Images](https://mlsanity.com/paper/2510.04225/) (2025)
- [Enhancing the Outcome Reward-based RL Training of MLLMs with Self-Consistency Sampling](https://mlsanity.com/paper/2511.10648/) (2025)
- [FASTopoWM: Fast-Slow Lane Segment Topology Reasoning with Latent World Models](https://mlsanity.com/paper/2507.23325/) (2025)
- [VideoPro: Adaptive Program Reasoning for Long Video Understanding](https://mlsanity.com/paper/2509.17743/) (2025)
- [Learning to Think Fast and Slow for Visual Language Models](https://mlsanity.com/paper/2511.16670/) (2025)
- [SSPO: Self-traced Step-wise Preference Optimization for Process Supervision and Reasoning Compression](https://mlsanity.com/paper/2508.12604/) (2025)
- [Beyond Perceptual Shortcuts: Causal-Inspired Debiasing Optimization for Generalizable Video Reasoning in Lightweight MLLMs](https://mlsanity.com/paper/2605.01324/) (2026)
- [VisionPulse: Dynamic Visual Sparsity for Efficient Multimodal Reasoning](https://mlsanity.com/paper/2605.31457/) (2026)
- [Conditional Performance Guarantee for Large Reasoning Models](https://mlsanity.com/paper/2601.22790/) (2026)
- [SesaHand: Enhancing 3D Hand Reconstruction via Controllable Generation with Semantic and Structural Alignment](https://mlsanity.com/paper/2603.00443/) (2026)
- [A Survey of Multimodal Mathematical Reasoning: From Perception, Alignment to Reasoning](https://mlsanity.com/paper/2603.08291/) (2026)

## 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.