# OPD-V: Visual On-Policy Self-Distillation with Modality Balance

arXiv: https://arxiv.org/abs/2608.05131 · HTML: https://mlsanity.com/paper/2608.05131/

## 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 39 of its 46 references that are in the map.
- First posted on arXiv 5 Aug 2026.

## Placed in

- LLM reasoning / [Multimodal reasoning](https://mlsanity.com/lr/multimodal-reasoning/)
- Multimodal LLMs / vision-language models / [MLLM training and visual instruction tuning](https://mlsanity.com/mllm/mllm-training-and-visual-instruction-tuning/)
- Knowledge distillation / [Multimodal distillation](https://mlsanity.com/distill/multimodal-distillation/)

## Builds on (most cited first, 30 of 39)

- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Qwen3-VL Technical Report](https://mlsanity.com/paper/2511.21631/) (2025)
- [On-Policy Distillation of Language Models: Learning from Self-Generated Mistakes](https://mlsanity.com/paper/2306.13649/) (2023)
- [Sequence-Level Knowledge Distillation](https://mlsanity.com/paper/1606.07947/) (2016)
- [V*: Guided Visual Search as a Core Mechanism in Multimodal LLMs](https://mlsanity.com/paper/2312.14135/) (2023)
- [GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learning](https://mlsanity.com/paper/2507.01006/) (2025)
- [Self-Distilled Reasoner: On-Policy Self-Distillation for Large Language Models](https://mlsanity.com/paper/2601.18734/) (2026)
- [Policy Distillation](https://mlsanity.com/paper/1511.06295/) (2015)
- [DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning](https://mlsanity.com/paper/2505.14362/) (2025)
- [MME-RealWorld: Could Your Multimodal LLM Challenge High-Resolution Real-World Scenarios that are Difficult for Humans?](https://mlsanity.com/paper/2408.13257/) (2024)
- [MiniCPM-V 4.5: Cooking Efficient MLLMs via Architecture, Data, and Training Recipe](https://mlsanity.com/paper/2509.18154/) (2025)
- [When More is Less: Understanding Chain-of-Thought Length in LLMs](https://mlsanity.com/paper/2502.07266/) (2025)
- [Thyme: Think Beyond Images](https://mlsanity.com/paper/2508.11630/) (2025)
- [Divide, Conquer and Combine: A Training-Free Framework for High-Resolution Image Perception in Multimodal Large Language Models](https://mlsanity.com/paper/2408.15556/) (2024)
- [DeepEyesV2: Toward Agentic Multimodal Model](https://mlsanity.com/paper/2511.05271/) (2025)
- [Don't Overthink it. Preferring Shorter Thinking Chains for Improved LLM Reasoning](https://mlsanity.com/paper/2505.17813/) (2025)
- [Vision-OPD: Learning to See Fine Details for Multimodal LLMs via On-Policy Self-Distillation](https://mlsanity.com/paper/2605.18740/) (2026)
- [PRISM: Self-Pruning Intrinsic Selection Method for Training-Free Multimodal Data Selection](https://mlsanity.com/paper/2502.12119/) (2025)
- [LLaVA Steering: Visual Instruction Tuning with 500x Fewer Parameters through Modality Linear Representation-Steering](https://mlsanity.com/paper/2412.12359/) (2024)
- [CoT-Kinetics: A Theoretical Modeling Assessing LRM Reasoning Process](https://mlsanity.com/paper/2505.13408/) (2025)
- [Zooming without Zooming: Region-to-Image Distillation for Fine-Grained Multimodal Perception](https://mlsanity.com/paper/2602.11858/) (2026)
- [ASCD: Attention-Steerable Contrastive Decoding for Reducing Hallucination in MLLM](https://mlsanity.com/paper/2506.14766/) (2025)
- [Evaluating and Steering Modality Preferences in Multimodal Large Language Model](https://mlsanity.com/paper/2505.20977/) (2025)
- [SenseNova-MARS: Empowering Multimodal Agentic Reasoning and Search via Reinforcement Learning](https://mlsanity.com/paper/2512.24330/) (2025)
- [Visual-Advantage On-Policy Distillation for Vision-Language Models](https://mlsanity.com/paper/2605.21924/) (2026)
- [Self-Evolving Multi-Agent Systems via Textual Backpropagation](https://mlsanity.com/paper/2506.09046/) (2025)
- [SPOT! Revisiting Video-Language Models for Event Understanding](https://mlsanity.com/paper/2311.12919/) (2023)
- [Visual-OPSD: Cross-Modal On-Policy Self-Distillation for Efficient Unified Multimodal Reasoning](https://mlsanity.com/paper/2606.18974/) (2026)
- [Generalizing from SIMPLE to HARD Visual Reasoning: Can We Mitigate Modality Imbalance in VLMs?](https://mlsanity.com/paper/2501.02669/) (2025)

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

- [Asymptotic Risk Calibration for Selective Question Answering](https://mlsanity.com/paper/2608.12008/) (2026)
- [Evidence-Aligned Multimodal On-Policy Self-Distillation for Fine-Grained Visual Understanding](https://mlsanity.com/paper/2609.34672/) (2026)
- [PE-OPSD: Internalizing Prompt Enhancement into Flow-matching Models via On-Policy Self-Distillation](https://mlsanity.com/paper/2609.36638/) (2026)
- [On-Policy Visual Evidence Distillation](https://mlsanity.com/paper/2609.36838/) (2026)
- [EviViT: Evidence-Adaptive Vision Transformers for Fine-Grained Perception](https://mlsanity.com/paper/2609.37123/) (2026)
- [Is Better Teacher Supervision Enough? Unlocking Student-side Learning in Multimodal On-Policy Distillation](https://mlsanity.com/paper/2609.39120/) (2026)
- [Where-OPD: Spatially Guided On-Policy Self-Distillation of MLLMs with Synthetic Scenes](https://mlsanity.com/paper/2610.02117/) (2026)
- [Slow-Fast Multi-Teacher On-Policy Distillation for Capability Preservation](https://mlsanity.com/paper/2610.02324/) (2026)

## Abstract (arXiv)

On-Policy Self-Distillation (OPSD) has become a standard post-training approach for improving visual reasoning in multimodal large language models (MLLMs). Existing methods draw privileged information from diverse input sources to guide self-distillation. Yet these designs overlook Modality Imbalance, a challenge inherent to MLLM reasoning. When textual information dominates generation, the model cannot fully integrate its multimodal input. Consequently, carefully designed privileged information remains underused, limiting the effectiveness of OPSD. To examine this limitation, we construct a Positive Teacher with the Zoom-In Image and a Negative Teacher with the Mask Image, which exhibit different degrees of Modality Imbalance. Changes in their reasoning correctness and token logits reveal that Modality Balance can itself serve as privileged information. Motivated by this finding, we introduce OPD-V, a visual OPSD paradigm that instantiates such information through the Positive Teacher and Negative Teacher. Positive Modality-Balance Logits Margins define a Modality-Balance Trust Region that selects the on-policy tokens used for self-distillation. Experiments across 6 benchmarks, 4 MLLM backbones, and 5 post-training methods show that OPD-V consistently improves reasoning performance while reducing training cost.