Evaluating small language models using an agent framework to execute complex geospatial data science workflows. Performance is evaluated on a benchmark built from the Axis Geospatial’s Geospatial Agent Benchmarks methodology to measure spatial reasoning, tool use, and code accuracy.