Google expands Geospatial Reasoning agent access and releases two benchmarks

Google Research publishes a technical report on its experimental Geospatial Reasoning agent, which connects Gemini to Earth AI foundation models and planet-scale data.

The report compares four agent configurations across four benchmarks, and Google is releasing two of them, GeoKTools and GeoExpertQueries, for researchers to test their own geospatial reasoning agents. Google also announces expanded early access to Ask Google Earth deep analysis, the Planetary Prediction Engine, and an agent-to-agent API, with early partner testing and limited access for these tools.

Google Research says the agent, which pairs domain-specific foundation models, spatial tools, and upfront planning, can carry out complex spatial analysis and planetary prediction that standalone LLMs cannot. The reports do not give benchmark scores or the scope of the access expansion.