Geospatial Intelligence Index v2: Methodology

Pokkiri, Najah

Version 2 of the Geospatial Intelligence Index, current since 13 November 2026. It combines the Geospatial Agent Index (Earth-Bench V2, GeoBenchX and GS-QA2, with Vision agents in the Text + vision view), Vision, Guided workflows (GeoAgentBench and GeoAnalystBench) and Geospatial knowledge. The rules shared by every evaluation, the limitations and the data sources are on the methodology page.

Cite this index

Pokkiri, N. (2026). Geospatial Intelligence Index v2: Methodology. Axis Spatial. https://www.axisspatial.com/arena/methodology/intelligence-index

BibTeX:

@techreport{pokkiri2026gii,
  author      = {Pokkiri, Najah},
  title       = {Geospatial Intelligence Index v2: Methodology},
  institution = {Axis Spatial},
  year        = {2026},
  version     = {2.0},
  url         = {https://www.axisspatial.com/arena/methodology/intelligence-index},
}

Groups and weights

A weighted average of four groups. Each index has two views, Text only first: Text only leaves out the evaluations that need images and weighs the rest out of 6; Text + vision weighs all four groups out of 8.

Group	Text + vision	Text only	What it measures
Agent tasks	4/8	4/6	The model's Geospatial Agent Index score, in the matching view (Text only or Text + vision)
Vision	2/8	Not included	Questions answered from satellite and aerial images
Guided workflows	1/8	1/6	The average of the model's GeoAgentBench and GeoAnalystBench scores. They count here, not in the Geospatial Agent Index
Geospatial knowledge	1/8	1/6	Closed-book multiple-choice questions on GIS and Earth observation

Agent tasks carry half because they measure the work itself and leave the most room between models; Vision a quarter, because Earth observation is imagery; Guided workflows and Geospatial knowledge an eighth each, as checks of the basics. There is no general-knowledge group: all the knowledge we ask about is geospatial.

Retiring a saturated group

When 80% of ranked models score 90 or more on a group in two versions in a row, the group is retired or replaced in a new version: Guided workflows by workflow tasks that do not give the method, Geospatial knowledge by a reasoning set. In version 1.0, ranked models scoring 90 or more: on Guided workflows, 13 of 14 (93%) in the default view and 22 of 24 (92%) across every reasoning level; on Geospatial knowledge, 13 of 14 (93%) and 20 of 24 (83%). At least one share is at 80% or more; the rule applies once that holds in two versions in a row.

Evaluations

Agent tasks

Included in Geospatial Intelligence Index v2 at 4/8 weighting (Text + vision) and 4/6 (Text only).

The model's Geospatial Agent Index score, in the matching view (Text only or Text + vision). Its evaluations, harnesses, sandbox, scoring and trace review are on the Geospatial Agent Index methodology page.

Vision

Included in Geospatial Intelligence Index v2 at 2/8 weighting (Text + vision only).

136 questions answered from satellite and aerial images (50 from Earth-Bench V2, 86 from ThinkGeo) in one model call each, with no tools.

Scoring and rejected questions

Each answer is scored pass or fail by code against its key, and the score is the plain average of the two sources' scores. Questions the provider rejects for their input, such as more images than it accepts, are counted and left out of the score. Each model has one run for now, so a score can move by about 4 points between runs.

Geospatial knowledge

Included in Geospatial Intelligence Index v2 at 1/8 weighting (Text + vision) and 1/6 (Text only).

1,685 closed-book multiple-choice questions on GIS and Earth observation, one model call each at the model's index reasoning level, with no tools and no internet.

Answers and scoring

The model may reason, then gives its answer; code checks the option letters, with no judge. A multiple-answer question is right only when the model names exactly the right set. A question a provider error left unanswered is asked again; the score is the share of scored questions answered correctly. The questions measure recall of published knowledge, not the quality of the source project's own model. A reasoning set will be added later, averaged equally with it.

Guided workflows: GeoAgentBench and GeoAnalystBench

Included in Geospatial Intelligence Index v2 at 1/8 weighting (Text + vision) and 1/6 (Text only). Standalone evaluations, not part of Geospatial Agent Index v2.

GIS workflows where the task gives the method step by step and the model writes and runs the code: reprojection, overlay, buffering, raster algebra, terrain and networks, usually ending in data files and a map. Top entries score 92 to 100, so these tasks count in the Intelligence Index but not in the Geospatial Agent Index.

Reference results

GeoAgentBench's reference results come from rerunning each task's recorded toolchain with the source's own tools. GeoAnalystBench was built to test workflow and code generation; here the agent runs its workflow and is graded on the resulting data and maps.

Harnesses

Kimi models are scored in Kimi Code once their runs there are complete on the three text evaluations. Because the Intelligence Index scores each model in its own harness, part of a difference between two models can come from the harness; the Geospatial Agent Index is the place to compare harnesses.

Ranking

Models in the Geospatial Intelligence Index. A model is ranked in a view when every group of that view has a score: Agent tasks (its entry ranked on the Geospatial Agent Index, in the harness the model is scored in), Guided workflows (a score on each guided evaluation, in the same harness), Geospatial knowledge and, in the Text + vision view, Vision, with at least 75% of the view's work done, weighted as the groups are. Any other model is listed under "Other entries" with the scores it has. It never has a 0.

Reasoning levels. By default each model is shown once, at its highest ranked reasoning level; each model's page lists every level with its own rank, and the Models page lists every entry. Cost and output tokens per index task are those of one attempt at a Geospatial Agent Index task, in the harness the model is scored in.

Models that cannot take images

A model that cannot take images, according to its developer's documentation, has no Vision or Vision agents score and no Text + vision index: it reads "No image input (see Text only)" and is ranked in the Text only view. We never apply this because our serving route lacks image input. Where the documentation is unclear, the model is not yet ranked until it has a vision run. A model that can take images but has no Vision agents run at its own reasoning level in its own harness reads "Not run: Vision agents". Scores are never copied from another reasoning level, harness or method.

Funding and conflicts

Axis Spatial runs and pays for Geospatial Agent Benchmarks. Some runs use free model access or credits from model providers and compute programmes, which give them no say over results, the method or listing. Labs and providers cannot pay for results, methodology changes or listing. Axis Spatial also offers paid custom evaluations, which never change the public method or scores: see trust and independence.

Spotted an issue? Contact us.

Version history

v2, 13 November 2026 to current

Authors: Pokkiri, N.

Each model is scored in its lab's own agent harness where the lab has one, otherwise in Terminus-2, Terminal-Bench's open harness, our default (6 and 7 November 2026).
A model is ranked in a view when every group of the view has a score and at least 75% of the view's work is done, weighted as the groups are; before, only when every group was complete (26 and 27 October 2026).
Each model is shown at its best ranked reasoning level, in place of its highest level run (6 November 2026).
A reply with no text, no reasoning and no output tokens is a provider error, asked again, not a wrong answer, in Geospatial knowledge and Vision (27 October 2026).
The Agent tasks group follows the Geospatial Agent Index v2 (see its version history). Weights, groups and views are unchanged.

v1, 19 October 2026 to 13 November 2026