Visual-Seeker: Towards Visual-Native Multimodal Agentic Search via Active Visual Reasoning

2026-06-13

AI Summary

This work addresses the challenge of achieving visually grounded multi-hop cross-modal reasoning in complex open-world scenarios, where existing approaches often fall short due to their reliance on static images and purely textual evidence, leading to factual inaccuracies. To overcome this limitation, we propose Visual-Seeker, an active visual reasoning agent that pioneers a vision-native search paradigm. Visual-Seeker dynamically attends to fine-grained visual details and iteratively gathers visual evidence to enable deep multimodal search. Our approach integrates multimodal large language models, an active visual attention mechanism, and cross-modal evidence fusion, trained on 5K synthetically generated high-quality multimodal reasoning trajectories. Evaluated on five challenging multimodal search benchmarks, Visual-Seeker achieves state-of-the-art performance, significantly outperforming multiple closed-source models and demonstrating robust visual reasoning and search capabilities in real-world web environments.

Abstract

Multimodal large language models (MLLMs) have demonstrated impressive capabilities in many visual tasks, but they often struggle with factual grounding when confronted with complex, open-world scenarios. While recent multimodal deep search agents attempt to address this issue by utilizing external tools, the visual-native search paradigm remains underexplored. Existing methods primarily rely on simple images with explicit semantics and text-only evidence trajectories, limiting the agent's ability to perform multi-hop, cross-modal reasoning and search. To address these limitations, we propose Visual-Seeker, a visual-native multimodal deep search agent via active visual reasoning. Rather than treating vision as a static input, our agent actively attends to fine-grained visual details, dynamically harvests visual evidence throughout the search process. To unlock its visual-native potential, we design an active visual reasoning data pipeline and synthesize 5K high-quality multimodal trajectories for model training. Extensive experiments demonstrate the state-of-the-art performance across five challenging multimodal search benchmarks, even surpassing several proprietary models, validating robust visual-native reasoning and search in real-world web environments. The code and data can be accessed at: https://github.com/ZhengboZhang/Visual-Seeker.

Problem

• multimodal search
• visual reasoning
• factual grounding
• open-world scenarios
• cross-modal reasoning

Innovation

• visual-native search
• active visual reasoning
• multimodal agent
• evidence trajectory
• fine-grained visual attention