# Towards On-Policy Data Evolution for Visual-Native Multimodal Deep Search Agents

arXiv: https://arxiv.org/abs/2605.10832 · HTML: https://mlsanity.com/paper/2605.10832/

## 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 22 of its 32 references that are in the map.
- First posted on arXiv 11 May 2026.

## Placed in

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

## Builds on (most cited first, 22 of 22)

- [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)
- [Humanity's Last Exam](https://mlsanity.com/paper/2501.14249/) (2025)
- [V*: Guided Visual Search as a Core Mechanism in Multimodal LLMs](https://mlsanity.com/paper/2312.14135/) (2023)
- [DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning](https://mlsanity.com/paper/2505.14362/) (2025)
- [Thinking with Images for Multimodal Reasoning: Foundations, Methods, and Future Frontiers](https://mlsanity.com/paper/2506.23918/) (2025)
- [FVQA: Fact-based Visual Question Answering](https://mlsanity.com/paper/1606.05433/) (2016)
- [APIGen-MT: Agentic Pipeline for Multi-Turn Data Generation via Simulated Agent-Human Interplay](https://mlsanity.com/paper/2504.03601/) (2025)
- [DeepEyesV2: Toward Agentic Multimodal Model](https://mlsanity.com/paper/2511.05271/) (2025)
- [WebWatcher: Breaking New Frontier of Vision-Language Deep Research Agent](https://mlsanity.com/paper/2508.05748/) (2025)
- [MMSearch: Benchmarking the Potential of Large Models as Multi-modal Search Engines](https://mlsanity.com/paper/2409.12959/) (2024)
- [AgentInstruct: Toward Generative Teaching with Agentic Flows](https://mlsanity.com/paper/2407.03502/) (2024)
- [Visual Agentic Reinforcement Fine-Tuning](https://mlsanity.com/paper/2505.14246/) (2025)
- [Facilitating Multi-turn Function Calling for LLMs via Compositional Instruction Tuning](https://mlsanity.com/paper/2410.12952/) (2024)
- [MMSearch-Plus: Benchmarking Provenance-Aware Search for Multimodal Browsing Agents](https://mlsanity.com/paper/2508.21475/) (2025)
- [Vision-DeepResearch Benchmark: Rethinking Visual and Textual Search for Multimodal Large Language Models](https://mlsanity.com/paper/2602.02185/) (2026)
- [Synthesizing Post-Training Data for LLMs through Multi-Agent Simulation](https://mlsanity.com/paper/2410.14251/) (2024)
- [AgentVista: Evaluating Multimodal Agents in Ultra-Challenging Realistic Visual Scenarios](https://mlsanity.com/paper/2602.23166/) (2026)
- [CoEvolve: Training LLM Agents via Agent-Data Mutual Evolution](https://mlsanity.com/paper/2604.15840/) (2026)
- [OpenSearch-VL: An Open Recipe for Frontier Multimodal Search Agents](https://mlsanity.com/paper/2605.05185/) (2026)
- [LoopTool: Closing the Data-Training Loop for Robust LLM Tool Calls](https://mlsanity.com/paper/2511.09148/) (2025)
- [Vision Search Assistant: Empower Vision-Language Models as Multimodal Search Engines](https://mlsanity.com/paper/2410.21220/) (2024)

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

- [SearchEyes: Towards Frontier Multimodal Deep Search Intelligence via Search World Simulation](https://mlsanity.com/paper/2607.05943/) (2026)
- [PhoneBuddy: Training Open Models for Agentic Phone Use](https://mlsanity.com/paper/2606.23049/) (2026)
- [What Makes Good Agentic Data? An ACE Lens on Data Generation for LLM Agents](https://mlsanity.com/paper/2608.27260/) (2026)
- [DeepSearch-World: Self-Distillation for Deep Search Agents in a Verifiable Environment](https://mlsanity.com/paper/2607.07820/) (2026)
- [V-Gym: Enhancing Agentic Visual Reasoning via Skill-Data Co-Evolution](https://mlsanity.com/paper/2609.34682/) (2026)
- [Towards Reliable AI Data Scientists: Data Agents with Workflow Harnesses](https://mlsanity.com/paper/2609.35255/) (2026)
- [EviScene: Beyond "Is It AI?" for Agentic Image Factuality Assessment](https://mlsanity.com/paper/2610.14609/) (2026)
- [Neos: Learning to Interleave Search and Visual Reasoning](https://mlsanity.com/paper/2610.17043/) (2026)

## Abstract (arXiv)

Multimodal deep search requires an agent to solve open-world problems by chaining search, tool use, and visual reasoning over evolving textual and visual context. Two bottlenecks limit current systems. First, existing tool-use harnesses treat images returned by search, browsing, or transformation as transient outputs, so intermediate visual evidence cannot be re-consumed by later tools. Second, training data is usually built by fixed curation recipes that cannot track the target agent's evolving capability. To address these challenges, we first introduce a visual-native agent harness centered on an image bank reference protocol, which registers every tool-returned image as an addressable reference and makes intermediate visual evidence reusable by later tools. On top of this harness, On-policy Data Evolution (ODE) runs a closed-loop data generator that refines itself across rounds from rollouts of the policy being trained. This per-round refinement makes each round's data target what the current policy still needs to learn. The same framework supports both diverse supervised fine-tuning data and policy-aware reinforcement learning data curation, covering the full training lifecycle of the target agent. Across 8 multimodal deep search benchmarks, ODE improves the Qwen3-VL-8B agent from 24.9% to 39.0% on average, surpassing Gemini-2.5 Pro in standard agent-workflow setting (37.9%). At 30B, ODE raises the average score from 30.6% to 41.5%. Further analyses validate the effectiveness of image-bank reuse, especially on complex tasks requiring iterative visual refinement, while rollout-feedback evolution yields more grounded SFT traces and better policy-matched RL tasks than static synthesis.