Skip to main content

How well AI understands the physical world, at planetary scale

We run the Spatial Intelligence Index: AI models benchmarked on real geospatial work, with every score and cost published. Then we evaluate them on your own tasks.

Try:

The top model is rarely the one you need.

We benchmark 41 AI models on the same geospatial tasks. The best one is not always the one to use. A model nearly as good costs about 1/5 as much per task. An open-weights model you can run on your own servers scores 79 of 100 for about a quarter the cost. Which is right depends on your work.

  1. Claude Opus 5.5Best overall89 of 1001×reference
  2. GPT-6.1 SolNearly as good, for less86 of 100About 1/5of the best model’s cost
  3. GPT-6 LunaCheapest strong model80 of 100About 1/36of the best model’s cost
  4. DeepSeek V4 FlashBest-value open-weights model. You can run it on your own servers79 of 100About a quarterof the best model’s cost

What we do

See the Spatial Intelligence Index

41 AI models benchmarked on the same geospatial tasks, with the score, cost per task and failures for each. It is free to read.

See the index

Evaluate models on your own tasks

We run the models on tasks from your work, with your data, under the same checks as the Spatial Intelligence Index. You get the score, cost per task and failures for each.

Tell us about your tasks

For AI labs: evaluation tasks, training data and environments

We write geospatial evaluation tasks, training data and environments, each with a verifier that checks the result.

Tell us what you need

Applied work

Need it built? Axis Agents builds the agents and workflows, and moves desktop GIS work to the cloud with axis-migrate. Models are chosen using the index.

Go to Axis Agents

Choose a model on your own tasks.

Send us one task your team does today and the data it uses. We evaluate the models on it and send back the score, cost per task and failures for each.

Work with us