Benchmark RadarChoroplethMap-Bench
ChoroplethMap-Bench evaluates spatial understanding of foundation models with 2,400 synthetic choropleth maps, corresponding GeoJSON data, and 12,000 questions across five cognitive dimensions (Identify, Spatial Recognition, Compare, Rank, Delineate). Models are assessed under Data Only, Map Only, and Data + Map conditions.
- Released
- 2026-07-20
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
The benchmark assesses whether cartographic representations add value over structured geodata for machine spatial reasoning, addressing a gap in evaluating map-based inputs. It supports decisions on when to incorporate visual map data in geospatial AI systems.
Motivation
Spatial understanding is crucial for foundation models (FMs), and maps have long helped humans organize and reason about geographic information.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.