TerraBench: Can Agents Reason Over Heterogeneous Earth-System Data?
Researchers have developed TerraBench, a new benchmark designed to evaluate the reasoning capabilities of AI agents when dealing with complex Earth-system data. The benchmark is built upon TerraAgent, a framework that combines large language models with scientific tools for data retrieval, geospatial processing, and simulation. TerraBench aims to address the limitations of current AI models in handling heterogeneous data sources crucial for climate and environmental decision-making. AI
IMPACT This benchmark could accelerate the development of AI agents capable of complex scientific reasoning over diverse environmental datasets.