# Semi-off-Policy Reinforcement Learning for Vision-Language Slow-Thinking Reasoning

arXiv: https://arxiv.org/abs/2507.16814 · HTML: https://mlsanity.com/paper/2507.16814/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- Placed in Multimodal reasoning, one of 17 directions in LLM reasoning.
- Builds on 36 of its 72 references that are in the map.
- First posted on arXiv 22 Jul 2025.

## Placed in

- LLM reasoning / [Multimodal reasoning](https://mlsanity.com/lr/multimodal-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, 30 of 36)

- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [DAPO: An Open-Source LLM Reinforcement Learning System at Scale](https://mlsanity.com/paper/2503.14476/) (2025)
- [Evaluating Object Hallucination in Large Vision-Language Models](https://mlsanity.com/paper/2305.10355/) (2023)
- [Tulu 3: Pushing Frontiers in Open Language Model Post-Training](https://mlsanity.com/paper/2411.15124/) (2024)
- [Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?](https://mlsanity.com/paper/2504.13837/) (2025)
- [SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training](https://mlsanity.com/paper/2501.17161/) (2025)
- [Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models](https://mlsanity.com/paper/2503.06749/) (2025)
- [Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs](https://mlsanity.com/paper/2412.21187/) (2024)
- [Visual-RFT: Visual Reinforcement Fine-Tuning](https://mlsanity.com/paper/2503.01785/) (2025)
- [LLaVA-CoT: Let Vision Language Models Reason Step-by-Step](https://mlsanity.com/paper/2411.10440/) (2024)
- [R1-Onevision: Advancing Generalized Multimodal Reasoning through Cross-Modal Formalization](https://mlsanity.com/paper/2503.10615/) (2025)
- [A Survey on Hallucination in Large Vision-Language Models](https://mlsanity.com/paper/2402.00253/) (2024)
- [Analyzing and Mitigating Object Hallucination in Large Vision-Language Models](https://mlsanity.com/paper/2310.00754/) (2023)
- [MM-Eureka: Exploring the Frontiers of Multimodal Reasoning with Rule-based Reinforcement Learning](https://mlsanity.com/paper/2503.07365/) (2025)
- [Towards Large Reasoning Models: A Survey of Reinforced Reasoning with Large Language Models](https://mlsanity.com/paper/2501.09686/) (2025)
- [Enhancing the Reasoning Ability of Multimodal Large Language Models via Mixed Preference Optimization](https://mlsanity.com/paper/2411.10442/) (2024)
- [Mulberry: Empowering MLLM with o1-like Reasoning and Reflection via Collective Monte Carlo Tree Search](https://mlsanity.com/paper/2412.18319/) (2024)
- [R1-Zero's "Aha Moment" in Visual Reasoning on a 2B Non-SFT Model](https://mlsanity.com/paper/2503.05132/) (2025)
- [Imitate, Explore, and Self-Improve: A Reproduction Report on Slow-thinking Reasoning Systems](https://mlsanity.com/paper/2412.09413/) (2024)
- [Teaching Large Language Models to Reason with Reinforcement Learning](https://mlsanity.com/paper/2403.04642/) (2024)
- [LLMs Can Easily Learn to Reason from Demonstrations Structure, not content, is what matters!](https://mlsanity.com/paper/2502.07374/) (2025)
- [RLAIF-V: Open-Source AI Feedback Leads to Super GPT-4V Trustworthiness](https://mlsanity.com/paper/2405.17220/) (2024)
- [Evaluation of OpenAI o1: Opportunities and Challenges of AGI](https://mlsanity.com/paper/2409.18486/) (2024)
- [Back to Basics: Revisiting REINFORCE Style Optimization for Learning from Human Feedback in LLMs](https://mlsanity.com/paper/2402.14740/) (2024)
- [O1 Replication Journey -- Part 2: Surpassing O1-preview through Simple Distillation, Big Progress or Bitter Lesson?](https://mlsanity.com/paper/2411.16489/) (2024)
- [WildVision: Evaluating Vision-Language Models in the Wild with Human Preferences](https://mlsanity.com/paper/2406.11069/) (2024)
- [Mitigating Object Hallucinations in Large Vision-Language Models with Assembly of Global and Local Attention](https://mlsanity.com/paper/2406.12718/) (2024)

## Built on by (most cited first, 8 of 8)

- [Intern-S1: A Scientific Multimodal Foundation Model](https://mlsanity.com/paper/2508.15763/) (2025)
- [Dynamic Skill Lifecycle Management for Agentic Reinforcement Learning](https://mlsanity.com/paper/2605.10923/) (2026)
- [Seeing with You: Perception-Reasoning Coevolution for Multimodal Reasoning](https://mlsanity.com/paper/2603.28618/) (2026)
- [OMIBench: Benchmarking Olympiad-Level Multi-Image Reasoning in Large Vision-Language Model](https://mlsanity.com/paper/2604.20806/) (2026)
- [RSICCLLM: A Multimodal Large Language Model for Remote Sensing Image Change Captioning](https://mlsanity.com/paper/2606.28266/) (2026)
- [Progressive Agent Skill Generation via Reinforcement Learning](https://mlsanity.com/paper/2608.01678/) (2026)
- [CPPO: Contrastive Perception Policy Optimization for VLM Agents](https://mlsanity.com/paper/2601.00501/) (2026)
- [Learning Multimodal Embeddings with Evidence-Aligned Readout](https://mlsanity.com/paper/2609.33659/) (2026)

## Abstract (arXiv)

Enhancing large vision-language models (LVLMs) with visual slow-thinking reasoning is crucial for solving complex multimodal tasks. However, since LVLMs are mainly trained with vision-language alignment, it is difficult to adopt on-policy reinforcement learning (RL) to develop the slow thinking ability because the rollout space is restricted by its initial abilities. Off-policy RL offers a way to go beyond the current policy, but directly distilling trajectories from external models may cause visual hallucinations due to mismatched visual perception abilities across models. To address these issues, this paper proposes SOPHIA, a simple and scalable Semi-Off-Policy RL for vision-language slow-tHInking reAsoning. SOPHIA builds a semi-off-policy behavior model by combining on-policy visual understanding from a trainable LVLM with off-policy slow-thinking reasoning from a language model, assigns outcome-based rewards to reasoning, and propagates visual rewards backward. Then LVLM learns slow-thinking reasoning ability from the obtained reasoning trajectories using propagated rewards via off-policy RL algorithms. Extensive experiments with InternVL2.5 and InternVL3.0 with 8B and 38B sizes show the effectiveness of SOPHIA. Notably, SOPHIA improves InternVL3.0-38B by 8.50% in average, reaching state-of-the-art performance among open-source LVLMs on multiple multimodal reasoning benchmarks, and even outperforms some closed-source models (e.g., GPT-4.1) on the challenging MathVision and OlympiadBench, achieving 49.08% and 49.95% pass@1 accuracy, respectively. Analysis shows SOPHIA outperforms supervised fine-tuning and direct on-policy RL methods, offering a better policy initialization for further on-policy training.