Axis Spatial Spatial Intelligence Index

Search modelsMenu

Spatial reasoning

A spatial intelligence benchmark for AI models: finding a place from its name or its coordinates, saying what lies within, next to or nearest to it, measuring distance and direction, and reading how a place changed across years of imagery.

Best score on each set

Best score any model reaches on each set, 0 to 100, hardest first. Each set opens its own page, with every model.

Saturated at the top: the best model already scores 100, so the set no longer separates the leading models.

What it covers

Some sets are answered by an agent with code tools and a spatial database, some from satellite and aerial images alone, and some from the model's own knowledge with no tools. Every model gets the same questions and limits and is scored the same way.

Every evaluation · Methodology

Frequently asked questions

What is a spatial intelligence benchmark?
A benchmark of how well an AI model understands the physical world: where places are, how they relate to each other, and how they change. Spatial reasoning is the part that works on locations and relations, as distinct from recalling facts about them.
Do these sets count in an index?
Five of them do: DVL-Bench, BigEarthNet.txt, GeoChrono, MultiGlobeQA and GS-QA2 Full make up the Spatial Intelligence Index. Each set has its own page with every model's score. The Geospatial Intelligence Index and the Geospatial Agent Index rank models on other work.