Google has unveiled significant new capabilities for its Earth AI platform, introducing a Geospatial Reasoning agent capable of conducting complex multi-step analysis across planet-scale datasets. In a technical report released October 14, the company demonstrated how combining specialized foundation models with spatial tools and upfront planning enables the agent to solve problems that generic large language models cannot handle alone. The agent integrates Google's suite of geospatial models—spanning remote sensing, weather forecasting, flood hydrology, and population dynamics—alongside tools from Google Maps Platform, Earth Engine, and Data Commons to tackle real-world spatial challenges. The company has been testing the agent with a select group of social impact and commercial partners, including the World Health Organization's Africa Regional Office, disaster relief organizations, and energy companies. Real-world applications demonstrated include disease outbreak investigations, disaster preparedness planning, and clean energy infrastructure assessment. Google reported that combining population and satellite signals improved health outcome predictions by 12 percent over demographic baselines while enabling mapping at eight times finer resolution than standard methods. To democratize access, Google is expanding experimental availability through two channels: a deep research update for its Ask Google Earth tool and an Agent-to-Agent API for developers building custom workflows. The company is also releasing two new benchmarks—GeoKTools and GeoExpertQueries—to help researchers evaluate their own geospatial reasoning systems. The platform now accepts user-provided data via CSV, GeoJSON, or cloud tables, allowing organizations to combine their local datasets with Earth AI's global models for customized analysis and autonomous training of predictive models.