Google Earth AI Geospatial Reasoning Agent: Capabilities and a Systematic Evaluation
Noam Slonim, Shravya Shetty and colleagues (Google Research) present the Geospatial Reasoning Agent (GRA) of Google Earth AI, a Gemini-based agent that organizes geospatial data, foundation models and tools into domain skills, plus two benchmarks to evaluate it.
Ask this paper
Design. GRA wraps Earth Engine data, geospatial foundation models and analysis tools as modular skills, and plans multi-stage workflows before executing them.
Benchmarks. GeoExpertQueries pairs practitioner problems with a consensus workflow from three experts and a human-validated autorater. GeoKTools uses templated queries with deterministic solvers that vary how many tool types are needed. UnivEARTH is found to be saturated.
Where the gain comes from. Comparing a bare LLM, an LLM with basic tools, GRA without planning, and full GRA shows the largest step comes from the skills and tools. The bare LLM answers every question without doing any computation.
Planning helps the smaller model. Upfront planning and execution guidance add 10.5 points on GeoExpertQueries with Gemini 3.8 Flash (52.2 to 62.7), but cost 3.2 points with Gemini 3.1 Pro.
Cost. Inside GRA, Gemini 3.8 Flash beats Gemini 3.1 Pro on all four benchmarks while costing 38% to 71% less, for example 62.7 against 47.6 on GeoExpertQueries at $0.199 against $0.688 per query.
Abstract
Google Earth AI is a family of geospatial AI models, together with an agent that orchestrates them. We present this agent, Google Earth AI's Geospatial Reasoning agent (GRA), a Gemini-powered research system that combines geospatial data sources, foundation models, and tools to answer questions about our planet. GRA organizes these into modular domain skills and pairs them with upfront geospatial planning to decompose complex queries into multi-stage workflows. Evaluating such systems is hard: real-world problems admit multiple valid workflows, require data discovery across large catalogs, and rely on evolving ground truth. We introduce two benchmarks and release their development splits. GeoExpertQueries poses complex analytical problems grounded in practitioner use cases and refined by geospatial experts; each is paired with a consensus golden workflow from three independent experts and scored by a human-validated, tool-agnostic autorater. GeoKTools consists of templated queries with deterministic reference solvers that systematically vary the number of required tool types. We also use two external benchmarks, CBGB and UnivEARTH; the latter we find to be saturated. To isolate the contribution of GRA's tools and its agentic orchestration, we compare four configurations, each with Gemini 3.1 Pro and 3.8 Flash: a standalone LLM, the LLM with basic tools, the LLM with GRA's skills and tools but without its upfront planning and execution guidance, and GRA itself. With at most basic tools, an advanced model recovers much of the answer on templated queries but satisfies only a small fraction of the expert workflows on open-ended problems; GRA performs well on both. The largest step comes from GRA's skills and tools; on open-ended problems, its planning and execution guidance add a further gain with Gemini 3.8 Flash, letting Flash match or exceed Pro at lower cost.