Researchers have developed MM-ToolSandBox, a framework for evaluating visual tool-calling agents. This benchmark and evaluation tool includes over 500 tools across 16 application domains and can handle multi-image, multi-turn tasks with realistic conversational phenomena. The framework generates diverse scenarios through information-flow-guided planning and quality filtering. Evaluations of state-of-the-art models show that they still lack robust visual tool-calling capability, with even the best model achieving a below 50% success rate. --- Why it matters: This matters to researchers in AI because it highlights the limitations of current models in visually grounded tasks and suggests fundamentally different research directions for improving models at different capability levels. Understanding these limitations is crucial for developing more robust visual tool-calling agents. Source: https://arxiv.org/abs/2607.11818