Researchers have developed a tool called CharTool that helps large language models better understand charts and graphs. The tool integrates with external tools to improve the model's ability to visually ground and perform precise numerical computations on chart data. To train CharTool, the researchers created DuoChart, a dataset of synthesized and real-world charts. Experiments showed that CharTool outperformed baseline models in several benchmarks, including one where it achieved competitive performance with larger or proprietary models. --- Why it matters: CharTool matters because it addresses a significant challenge in AI research: enabling large language models to accurately understand and reason about visual data like charts and graphs. This is an important problem for applications like scientific literature analysis and financial reporting, where accurate chart interpretation can have real-world consequences. Source: https://arxiv.org/abs/2604.02794