OpenVisTool-Bench spans 559 visual tool tasks

OpenVisTool-Bench packages 559 examples across five visual domains to evaluate multimodal agents that can invoke inspection and editing tools. The dataset spans charts, tables, visual search, GUI grounding, and web-to-HTML tasks; tool/no-tool gain metadata appears on the chart and table subsets. Builders extending agent discovery beyond a static catalog of tools should file this as a compact routing probe, not a definitive leaderboard: many images come from existing benchmarks and the card does not publish a contamination audit.