Towards Long-horizon Agentic Multimodal Search

arXiv cs.CV / 4/15/2026

📰 NewsSignals & Early TrendsIdeas & Deep AnalysisTools & Practical UsageModels & Research

共有:

Key Points

The paper proposes LMM-Searcher, a long-horizon multimodal deep search framework that reduces multimodal context explosion and token costs via file-based storage of visual assets referenced by lightweight UIDs.
It introduces a tailored fetch-image tool to load visual content on demand during active perception, enabling progressive, memory-efficient multimodal retrieval over many turns.
The authors build a data synthesis pipeline that creates queries requiring complex cross-modal multi-hop reasoning, then distill 12K trajectories to fine-tune Qwen3-VL-Thinking-30A3B into a specialized long-horizon search agent.
Experiments on four benchmarks show the approach scales to 100-turn search horizons and achieves state-of-the-art results among open-source models on tasks like MM-BrowseComp and MMSearch-Plus, with good generalizability across base models.
The authors indicate that the code will be released publicly at the provided GitHub repository link.

Abstract

Multimodal deep search agents have shown great potential in solving complex tasks by iteratively collecting textual and visual evidence. However, managing the heterogeneous information and high token costs associated with multimodal inputs over long horizons remains a critical challenge, as existing methods often suffer from context explosion or the loss of crucial visual signals. To address this, we propose a novel Long-horizon MultiModal deep search framework, named LMM-Searcher, centered on a file-based visual representation mechanism. By offloading visual assets to an external file system and mapping them to lightweight textual identifiers (UIDs), our approach mitigates context overhead while preserving multimodal information for future access. We equip the agent with a tailored fetch-image tool, enabling a progressive, on-demand visual loading strategy for active perception. Furthermore, we introduce a data synthesis pipeline designed to generate queries requiring complex cross-modal multi-hop reasoning. Using this pipeline, we distill 12K high-quality trajectories to fine-tune Qwen3-VL-Thinking-30A3B into a specialized multimodal deep search agent. Extensive experiments across four benchmarks demonstrate that our method successfully scales to 100-turn search horizons, achieving state-of-the-art performance among open-source models on challenging long-horizon benchmarks like MM-BrowseComp and MMSearch-Plus, while also exhibiting strong generalizability across different base models. Our code will be released in https://github.com/RUCAIBox/LMM-Searcher.