Google Expands Its Experimental Earth AI Geospatial Reasoning Agent

Google Research has published a technical report on its Earth AI Geospatial Reasoning agent, an experimental AI system that answers questions about the planet using maps, satellite data and forecasting models. Google announced on 14 October 2026 that it is also testing the agent with partners and widening early access.

What the agent does

Google describes the agent as “a Gemini-powered research system” that plans a multi-step analysis and then draws on Google’s geospatial tools and models, including its weather, flood and population models, Google Maps Platform and Earth Engine. Users can also bring their own data.

“In a new technical report, we demonstrate how combining domain-specific foundation models, spatial tools, and upfront planning enables a Geospatial Reasoning agent to carry out complex spatial analysis and planetary prediction that standalone LLMs can’t,” Google wrote.

Google’s test results

Google introduced two new benchmarks, GeoKTools and GeoExpertQueries, and tested four setups, from a plain language model up to the full agent with planning, using its Gemini 3.8 Flash and Gemini 3.1 Pro models. According to Google, “equipping the agent with Earth AI skills drives a significant jump across GeoExpertQueries, GeoKTools, and CBGB.” On one other benchmark, UnivEARTH, Google said even a standalone model scored 93.6%, likely because the answers are available on the public web. These are Google’s own results.

Google also noted a limit: its GeoExpertQueries scores are judged by an automated rater rather than by experts. “We therefore consider the agent’s output as a strong first draft for experts to review and build on, not a final answer,” it wrote.

Partner tests

Google said it has been testing the agent with organisations including the World Health Organization’s Regional Office for Africa, People in Need, GiveDirectly, the Milken Institute and WPP. In one example Google gave, WHO’s Africa team used the agent at the start of the 2026 Ebola outbreak in the Democratic Republic of the Congo. “In minutes, the team pinpointed 48 exposed settlements and located more than 45,500 at-risk people — a process that normally would have taken weeks,” Google said. These accounts come from Google and have not been independently checked.

How to get access

Google is widening experimental access in two ways: an experimental deep research update to Ask Google Earth, and an experimental Agent-to-Agent (A2A) service that lets other AI agents call it. Organisations can apply through an interest form.

What this means

• AI agents for the planet. Google wants its agent to do complex map-based analysis that chatbots can’t.
• Experts still in charge. Google calls the output a “first draft” for experts to check.
• Still experimental. Access is limited, and the results so far are Google’s own.

Sources

1. Google Research blog, “Putting Earth AI to work: Geospatial agents for our planet’s biggest challenges”, Shravya Shetty and Luke Barrington, 14 October 2026. https://research.google/blog/putting-earth-ai-to-work-geospatial-agents-for-our-planets-biggest-challenges/
2. Google Research, “Google Earth AI Geospatial Reasoning Agent: Capabilities and a Systematic Evaluation”, technical report, arXiv, 2026. https://research.google/pubs/google-earth-ai-geospatial-reasoning-agent-capabilities-and-a-systematic-evaluation/

Related stories

• Google DeepMind Gives Up to 200 European Researchers Access to Its Frontier Models, Plus $3M in Compute Credits
• Google Releases Gemini 4 Argon Worldwide: What It Costs, Who Gets It and What Google Claims
• Google’s Agent Development Kit Now Covers Five Languages: Python, Go, TypeScript, Java and Kotlin
• OpenAI Says AI Agents Now Account for More Than Half of Its EU Enterprise Use, in an Analysis of Its Own Customer Data