Overview
The Geospatial Reasoning agent of Google Earth AI is a system Google Research presented at its Research@ event on October 14, 2026. It combines domain foundation models, spatial tools and upfront planning to carry out multi-step spatial analysis that standalone large language models cannot.
The difficulty lies in the data itself: planetary datasets are massive, with specialised map and satellite formats that need careful multi-step planning to combine accurately. The agent brings four capabilities together, connecting Earth AI foundation models for remote sensing, weather, flood hydrology and population dynamics on one side, with Gemini handling orchestration and questions on the other.
The release comes in three parts: a technical report with systematic evaluation across four benchmarks, including GeoKTools and GeoExpertQueries, two new benchmarks Google developed and is releasing; insights from real-world partner testing; and expanded early access through Google Earth and Google Cloud, covering an experimental deep research update for Ask Google Earth and an agent-to-agent API for developers.
Key Features
- Connected to Earth foundation models: Taps Earth AI models for remote sensing, WeatherNext weather, flood hydrology and population dynamics
- Spatial tools with upfront planning: Plans multi-step operations over specialised map and satellite formats before executing them
- Systematic evaluation on four benchmarks: The technical report quantifies the value of specialised tools and planning
- Two new benchmarks released: GeoKTools and GeoExpertQueries are published for researchers to test their own geospatial agents
- Deep research in Ask Google Earth: Experimental multi-step geospatial analysis, open through early access
- Agent-to-agent API: Early access for developers building custom workflows, including Gemini integration recipes
- Planetary prediction engine: Generates risk assessments and forecasts across scenarios
- Tested in partner workflows: Work with the WHO Regional Office for Africa, People in Need, GiveDirectly, the Milken Institute and WPP
Use Cases
- Public health bodies investigating disease outbreaks and screening risk
- Humanitarian organisations preparing before a disaster hits
- Siting and screening for clean energy projects
- Cross-regional risk assessment and scenario forecasting for companies and institutions
- Developers wiring geospatial analysis into their own agent workflows
Pros
- Technical report and public benchmarks ship together, so the method can be checked
- Two new benchmarks let outside researchers compare their own agents
- Tested in real workflows at organisations including the WHO Regional Office for Africa
- Two entry points, Ask Google Earth and the A2A API, so either route is open
- Weather, remote sensing, hydrology and population models callable from one place
Pricing
Experimental capabilities open through early access, available via Ask Google Earth and a Google Cloud interest form, with agent-to-agent API early access on the developer side. Availability and billing follow the official Google pages.
Summary
The Geospatial Reasoning agent of Google Earth AI is a system Google Research presented on October 14, 2026, combining Earth foundation models, spatial tools and upfront planning for multi-step spatial analysis that general-purpose models cannot handle. It ships with a technical report covering four benchmarks, including GeoKTools and GeoExpertQueries, two benchmarks Google built and released. Partner testing covers the WHO Regional Office for Africa, People in Need, GiveDirectly, the Milken Institute and WPP across disease outbreak investigation, disaster preparedness and clean energy screening. Entry comes through two doors: an experimental deep research update in Ask Google Earth, and early access to an agent-to-agent API for developers with Gemini integration recipes. It fits organisations doing risk assessment, siting and scenario forecasting; for a static map, conventional GIS tooling is more direct.
Version History
- Geospatial reasoning technical report published with GeoKTools and GeoExpertQueries released as benchmarks (2026-10-14): On October 14, 2026, at its Research@ event, Google Research shared three updates on the Earth AI Geospatial Reasoning agent. First, a technical report showing that combining domain-specific foundation models, spatial tools and upfront planning enables complex spatial analysis and planetary prediction that standalone LLMs cannot, with systematic evaluations across four benchmarks including GeoKTools and GeoExpertQueries, two benchmarks Google developed and released for researchers to test their own geospatial reasoning agents. Second, insights from real-world partner testing over recent months with a small group of social impact and commercial organisations including the WHO Regional Office for Africa, People in Need, GiveDirectly, the Milken Institute and WPP, applying the agent to disease outbreak investigations, disaster preparedness and clean energy screening. Third, expanded early access: an experimental deep research update for Ask Google Earth covering multi-step geospatial analysis, and early access to an agent-to-agent API for developers building custom workflows, including integration recipes for connecting to Gemini.