ChartNet is a large-scale, high-quality multimodal dataset released in 2026 by MIT in collaboration with IBM Research and other institutions. The related research papers are as follows: ChartNet: A Million-Scale, High-Quality Multimodal Dataset for Robust Chart UnderstandingIt aims to address the shortcomings of existing models in joint reasoning of geometric visual patterns, structured numerical values, and text descriptions, and is widely used in the training and evaluation of graph understanding, cross-modal reasoning, and visual language models. This dataset contains 4.2 million synthetic chart samples, 94,643 manually verified chart samples, and 30,000 real-world charts, covering 24 chart types and 6 plotting libraries.

Dataset composition

• 4.2 million synthetic chart samples
• 94,643 examples of manually verified charts
• 2,000 manually verified assessment samples
• 30,000 real-world scene charts
• 24 types of charts
• 6 types of drawing libraries

Guidelines

• The core, reasoning, human_verified, grounded_qa, and full reasoning subsets comply with data availability notices and may not be used for commercial development or deployment.
• core_permissive subset adherence cdla-permissive-2.0 protocol