See or Say Graphs: Agent-Driven Scalable Graph Structure Understanding with Vision-Language Models

Vision-language models (VLMs) have shown promise in graph structure understanding, but remain limited by input-token constraints, facing scalability bottlenecks and lacking effective mechanisms to coordinate textual and visual modalities. To address these challenges, we propose GraphVista, a unified framework that enhances both scalability and modality coordination in graph structure understanding. For scalability, GraphVista organizes graph information hierarchically into a lightweight GraphRAG base, which retrieves only task-relevant textual descriptions and high-resolution visual subgraphs, compressing redundant context while preserving key reasoning elements. For modality coordination, GraphVista introduces a planning agent that decomposes and routes tasks to the most suitable modality-using the text modality for direct access to explicit graph properties and the visual modality for local graph structure reasoning grounded in explicit topology. Extensive experiments demonstrate that GraphVista scales to large graphs, up to 200× larger than those used in existing benchmarks, and consistently outperforms existing textual, visual, and fusion-based methods, achieving up to 4.4× quality improvement over the state-of-the-art baselines by fully exploiting the complementary strengths of both modalities.