Outperforming Larger Language Models
MIT and IBM Launch ChartNet to Improve AI Chart Analysis
New dataset helps vision-language models outperform GPT-4o by accurately reading axes, scales, and scientific data.
A digital display showing a complex scientific line graph with data points highlighted by a blue digital overlay, representing AI chart analysis.
Photo: Avantgarde News
Researchers from MIT and IBM launched ChartNet on June 3, 2026 [1]. This massive dataset helps vision-language models read complex charts more accurately [1][2]. The system focuses on interpreting axes, scales, and individual data points within scientific figures [1].
Early testing shows that small models trained on ChartNet outperform GPT-4o in data extraction [3]. This breakthrough addresses a major weakness in existing artificial intelligence systems [2]. The researchers aim to turn chart-reading into a rigorous test for future AI development [2].
Editorial notes
Transparency note
AI assisted drafting. Human edited and reviewed.
- AI assisted
- Yes
- Human review
- Yes
- Last updated
Risk assessment
Reviewed for sourcing quality and editorial consistency.
Sources
- 1.↗
news.mit.edu
https://news.mit.edu/2026/mit-researchers-teach-ai-models-to-interpret-charts-0603
- 2.↗
startupfortune.com
https://startupfortune.com/mits-chartnet-turns-chart-reading-into-a-serious-ai-test/
- 3.↗
otontechnology.com
https://otontechnology.com/mit-chartnet-dataset-small-model-beats-gpt-4o-charts/
Related stories
View all
Topics
About the author
Avantgarde News Desk covers outperforming larger language models and editorial analysis for Avantgarde News.