Google Research Unveils Geospatial Agent System Linking Gemini to Earth AI and Introduces Two New Benchmarks

Google Research has released a new geospatial reasoning agent system that connects Gemini to the Earth AI base model, enabling multi‑step reasoning and self‑training predictive models for planetary challenges. The announcement also includes the launch of two new benchmarks designed to evaluate the performance of such agents. The system builds on Google’s Earth AI foundation, allowing agents to interpret satellite imagery, generate hypotheses, and iteratively refine predictions. By integrating Gemini’s conversational capabilities, the agents can ask clarifying questions and provide more nuanced insights. The accompanying benchmarks aim to standardize evaluation across different geospatial tasks, fostering broader adoption and comparison of future models. This development marks a significant step toward automating complex Earth‑system analyses, though the long‑term impact on policy and decision‑making remains to be seen as researchers continue to test and refine the approach.

Written with AI from the reporting · updated 10/15/2026

1. 1Oct 15, 00:00recorded Oct 15, 22:26single sourcerevised

   Research@ 2026: 10 ways we're amplifying human ingenuity

   Read original source ↗ Google ResearchReplay this moment in the explorer
2. 2Oct 15, 00:00recorded Oct 15, 22:26single sourcerevised

   发布地理空间推理代理系统并发布新基准

   Read original source ↗ Google ResearchReplay this moment in the explorer

Uncertainty marks flag where to check the sources; they are not proven causation.

Different reports do not necessarily disagree. Check original sources and publication dates.