Research/Terra

An agent for complete
geospatial workflows.

Complete, auditable geospatial workflows using trusted processors, with reusable data published to catalogs built on open standards.

For the story behind Terra, how it works, and a walkthrough of the evaluation, read our introductory blog post.

Introducing Terra ↗GitHub repository ↗Original benchmark ↗

Agent architecture

The Terra harness

Terra builds and runs complete geospatial workflows using trusted processors, publishing reusable data into catalogs built on open standards. Every input, processing step and result is traceable—so the entire workflow can be inspected and audited.

Input

Analysis request

Spatial scope · required outputs

Terra

Agent harness

Output

Catalog datasets & map layers

Open standards · data provenance

01Investigation

Vertical-slice evaluation

Trial workflow

Representative spatial extent

Processor execution

Raster · vector · table

Artifact inspection

Statistics · geometry · previews

Plan revision

Resolve findings; revise the method

Workflow proposal

02Execution

Workflow execution

Saved workflow

Requested analysis scope

Processor execution

Raster · vector · table

Result assessment

Compare outputs with requirements

Workflow revision

Reuse suitable input artifacts

Simple Workflow Example

A recorded example

Chart contour lines for accumulated snow fall in winter 2023-2024 in USA

Actual executed workflow and step outcomes — Diagram generated from the saved workflow. This is not a map or proof of correctness; see independent checks.

Follow the recorded workflow and evidence ↗

Results at a glance

terra-geobenchx-v1

We built our evaluation around GeoBenchX, using its geospatial questions and datasets to test Terra’s complete agent system. Our evaluation framework runs the tasks through Terra and verifies the resulting data, maps and answers.

Computational tasks

115 / 133

86.5% passed

Verified end-to-end passes / valid measurements

Qualitative controls

67 / 68

98.5% passed

Rubric passes / valid control measurements

Evaluation coverage

201 / 202

Valid primary measurements / original questions · 202 dispositions recorded

Scores use each task’s first valid attempt, including partial results, failures, and timeouts. Setup and infrastructure issues are reported separately; retries don’t change the scores.

Some tests have known ambiguities in their instructions or grading. We include their original results in the reported scores for transparency and reproducibility.

Pass 182Partial 3Fail 15Timeout 1Setup blocked 1Invalid measurement 0

Median end-to-end duration: 14m 32s · 201/202 recorded
Total tokens (recorded tasks): 395,667,224 · 201/202

Timings include the complete recorded attempt, including timeouts. Usage covers only tasks with recorded measurements.

Published GeoBenchX match rates for eight models, split into solvable and unsolvable tasks. Exact values are available in the table below. — Solvable tasks measure successful solutions; unsolvable tasks measure correct rejections. Source: Krechetova & Kochedykov, GeoBenchX, Table 7 (arXiv v2, October 2025).

Published reference-solution match rates
Model	Solvable	Unsolvable
Gemini 2.0 Flash	39%	40%
Gemini 2.5 Pro Preview	39%	55%
Claude Haiku 3.5	40%	22%
Claude Sonnet 3.5	47%	47%
Claude Sonnet 4	55%	17%
GPT-4o	38%	62%
GPT-4.1	42%	43%
o4-mini	51%	90%

Explore the tasks

Outcomes & evidence

Computational work and identifying missing evidence are different capabilities. We report them separately, with the original task identity and the reason for each outcome.

202 of 202 tasks

Task	Type	Outcome	Duration	Evidence
How many Latin American seaports are located in regions with high population growth104119 / country-change-selection Evaluation notes Approval policy: explicit reduced or unavailable work is not preauthorized.; No accepted final answer was published. 61 tool calls	Computation	Fail	21m 19s	View task ↗
Create a heatmap showing railway station density in areas with high snow accumulation in the USA112207 / geographic-heatmap Evaluation notes Final count or unknown-coverage count disagrees with the computed evidence. 62 tool calls	Computation	Partial	19m 51s	View task ↗
Map drought severity indices for agricultural regions in Southern Africa?117526 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	2m 9s	View task ↗
Chart contour lines for accumulated snow fall in winter 2023-2024 in USA118857 / contour-map Evaluation notes Independent artifact and delivery checks passed. 42 tool calls	Computation	Pass	10m 25s	View task ↗
Identify regions in Brazil where railway density correlates with population density123067 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 20s	View task ↗
Visualize the unemployment rates in European Union member states.128902 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 27s	View task ↗
Lakes near Canada's mapped boundary129699 / water-feature-selection Evaluation notes Independent artifact and delivery checks passed. 53 tool calls	Computation	Pass	22m 2s	View task ↗
Facilities in negative-migration countries130168 / spatial-analysis Evaluation notes Independent artifact and delivery checks passed. 63 tool calls	Computation	Pass	31m 53s	View task ↗
Generate contour lines of snow accumulation near major US lakes131687 / contour-map Evaluation notes Independent artifact and delivery checks passed. 56 tool calls	Computation	Pass	19m 6s	View task ↗
Create a heatmap of earthquake occurrences in regions with high population growth133682 / geographic-heatmap Evaluation notes Heatmap units do not match the declared weights per cell. 70 tool calls	Computation	Fail	35m 12s	View task ↗
Calculate urban sprawl rates for Chinese cities between 2015-2020?138450 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 8 tool calls	Limitation control	Pass	1m 44s	View task ↗
Generate contour lines of population density in Angola's northern regions146485 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 4 tool calls	Limitation control	Pass	2m 5s	View task ↗
Create a heatmap showing population density in Chile's coastal areas.149968 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 3 tool calls	Limitation control	Pass	2m 10s	View task ↗
Lakes near observed deep-snow areas150069 / water-feature-selection Evaluation notes Independent artifact and delivery checks passed. 37 tool calls	Computation	Pass	25m 29s	View task ↗
US towns near both lakes and rivers154613 / water-feature-selection Evaluation notes Independent artifact and delivery checks passed. 43 tool calls	Computation	Pass	21m 57s	View task ↗
Show water withdrawal patterns in Himalayan nations169833 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 44 tool calls	Computation	Pass	14m 58s	View task ↗
Show population density patterns across regions170096 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 70 tool calls	Computation	Pass	39m 0s	View task ↗
Compare agricultural GDP contribution across Andean countries186506 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 57 tool calls	Computation	Pass	27m 9s	View task ↗
Show water withdrawal patterns in BRICS nations191153 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 35 tool calls	Computation	Pass	13m 40s	View task ↗
Calculate the percentage of Peru's population living more than 100km from seaports197777 / population-distance-share Evaluation notes Independent artifact and delivery checks passed. 58 tool calls	Computation	Pass	25m 12s	View task ↗
Compare the population density along railways in Bangladesh with the national average208110 / population-distance-share Evaluation notes Independent artifact and delivery checks passed. 63 tool calls	Computation	Pass	38m 10s	View task ↗
Create a map of urban population density in major cities.214430 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 3 tool calls	Limitation control	Pass	2m 2s	View task ↗
Create a heatmap of mineral facility density in African countries with large labor forces217200 / geographic-heatmap Evaluation notes Independent artifact and delivery checks passed. 77 tool calls	Computation	Pass	39m 48s	View task ↗
Map industrial pollution hotspots in Eastern European cities?221611 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 3 tool calls	Limitation control	Pass	1m 12s	View task ↗
How does electric consumption vary across Arctic nations?222041 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 42 tool calls	Computation	Pass	21m 5s	View task ↗
Show water withdrawal patterns in South Asian peninsula227822 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 46 tool calls	Computation	Pass	19m 33s	View task ↗
Display climate adaptation readiness across vulnerable nations.237711 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 3 tool calls	Limitation control	Pass	3m 3s	View task ↗
What is the relationship between population density and economic activity in flood-prone areas of Peru?244529 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 3 tool calls	Limitation control	Pass	2m 9s	View task ↗
Identify regions in Brazil where population density is in the bottom quartile nationally and are crossed by railways245463 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 7 tool calls	Limitation control	Pass	2m 48s	View task ↗
Visualize net migration patterns worldwide250712 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 58 tool calls	Computation	Pass	17m 45s	View task ↗
Make contour lines of snow accumulation for areas with railway stations in the USA251255 / contour-map Evaluation notes Independent artifact and delivery checks passed. 41 tool calls	Computation	Pass	14m 29s	View task ↗
Show the distribution of GDP per capita across different world regions252796 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 70 tool calls	Computation	Pass	22m 5s	View task ↗
Create a heatmap of seaport density in regions with high GDP growth in Latin America253892 / geographic-heatmap Evaluation notes Independent artifact and delivery checks passed. 69 tool calls	Computation	Pass	39m 20s	View task ↗
Show labor force distribution in Arctic Circle countries258248 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 35 tool calls	Computation	Pass	15m 21s	View task ↗
Create contour lines of snow accumulation in urban agglomerations264811 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	2m 3s	View task ↗
Create a heatmap of power station density in regions with high energy demand265245 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 45s	View task ↗
Show global labor force concentration270481 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 49 tool calls	Computation	Pass	11m 45s	View task ↗
Show the patterns of youth unemployment across Mediterranean countries.272016 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 56s	View task ↗
Make a heatmap of tuberculosis cases in New York State272736 / county-heatmap Evaluation notes Independent artifact and delivery checks passed. 67 tool calls	Computation	Pass	26m 31s	View task ↗
Compare rural population percentages in Western Asian countries275666 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 51 tool calls	Computation	Pass	30m 10s	View task ↗
Make a heatmap showing population concentration in flood-risk zones of South America279238 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 3 tool calls	Limitation control	Pass	2m 13s	View task ↗
Map the distribution of railway networks in relation to population density in Bangladesh.281727 / source-overlay Evaluation notes Independent artifact and delivery checks passed. 71 tool calls	Computation	Pass	22m 56s	View task ↗
Create contour lines from Angola's population data in relation to power stations291123 / contour-map Evaluation notes Independent artifact and delivery checks passed. 51 tool calls	Computation	Pass	15m 51s	View task ↗
Create a map showing forest coverage percentage across different countries296055 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 52 tool calls	Computation	Pass	19m 8s	View task ↗
What are main agriculture cultures in Blefuscu?299410 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 26s	View task ↗
Map the relationship between GDP per capita and electric power consumption per capita globally.301626 / bivariate-map Evaluation notes Independent artifact and delivery checks passed. 57 tool calls	Computation	Pass	24m 5s	View task ↗
Visualize water usage patterns in Great Lakes region of Africa304935 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 4 tool calls	Limitation control	Pass	2m 5s	View task ↗
How many power stations in Africa are located in economically strategic regions?305778 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 5 tool calls	Limitation control	Pass	2m 21s	View task ↗
Railway length inside observed deep-snow cells310610 / clipped-line-length Evaluation notes Independent artifact and delivery checks passed. 44 tool calls	Computation	Pass	36m 53s	View task ↗
Historical active wildfire incidents near Amtrak stations311281 / point-line-proximity Evaluation notes Independent artifact and delivery checks passed. 42 tool calls	Computation	Pass	19m 14s	View task ↗
Map air quality index for major cities in South Asia?315340 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 6 tool calls	Limitation control	Pass	1m 48s	View task ↗
High-density population and 2023–24 snowfall321268 / native-population-snow Evaluation notes Independent artifact and delivery checks passed. 48 tool calls	Computation	Pass	38m 23s	View task ↗
Algerian mines in dense population cells332083 / spatial-analysis Evaluation notes Independent artifact and delivery checks passed. 39 tool calls	Computation	Pass	12m 15s	View task ↗
Towns in deep snow333321 / spatial-analysis Evaluation notes Independent artifact and delivery checks passed. 40 tool calls	Computation	Pass	16m 56s	View task ↗
Generate a heatmap of railway density in high GDP regions of Brazil342420 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 11s	View task ↗
Generate a heatmap of USA population density and overlay it with current wildfire locations.345106 / source-overlay Evaluation notes Independent artifact and delivery checks passed. 57 tool calls	Computation	Pass	20m 52s	View task ↗
Map total forest area in square kilometers by region348947 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 56 tool calls	Computation	Pass	18m 11s	View task ↗
Compare forest coverage between Continental and Maritime Southeast Asia356076 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 48 tool calls	Computation	Pass	14m 32s	View task ↗
What is agriculture?364138 / qualitative-control Evaluation notes The accepted final answer never defines agriculture; it only reports an incomplete analysis and no verified results.; Treats a straightforward conceptual definition as blocked analytical work, despite no GIS observations being needed. 20 tool calls	Limitation control	Fail	5m 59s	View task ↗
River records intersecting highly forested South American countries365343 / forest-river-intersection Evaluation notes Independent artifact and delivery checks passed. 61 tool calls	Computation	Pass	33m 22s	View task ↗
Make a heatmap showing population concentration in earthquake-affected zones367674 / geographic-heatmap Evaluation notes Heatmap units do not match the declared weights per cell. 48 tool calls	Computation	Fail	17m 11s	View task ↗
Map the global distribution of GDP per capita375632 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 48 tool calls	Computation	Pass	11m 17s	View task ↗
Show digital government services adoption in developed economies.377506 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 18s	View task ↗
Calculate the total population affected by floods in Peru during February 2018381719 / observed-flood-population Evaluation notes Independent artifact and delivery checks passed. 65 tool calls	Computation	Pass	37m 36s	View task ↗
How does forest coverage vary across Pacific island nations?385630 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 53 tool calls	Computation	Pass	15m 30s	View task ↗
Map global rural-urban population distribution394050 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 63 tool calls	Computation	Pass	19m 9s	View task ↗
Generate a heatmap of population density around African mineral extraction sites397320 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 52s	View task ↗
Map the renewable water resources per capita in Middle Eastern countries399319 / country-choropleth Evaluation notes Approval policy: explicit reduced or unavailable work is not preauthorized.; No accepted final answer was published. 66 tool calls	Computation	Fail	23m 24s	View task ↗
Amtrak stations in earthquake counties417961 / spatial-analysis Evaluation notes Independent artifact and delivery checks passed. 54 tool calls	Computation	Pass	14m 47s	View task ↗
African power stations near railways418513 / point-line-proximity Evaluation notes Independent artifact and delivery checks passed. 37 tool calls	Computation	Pass	12m 42s	View task ↗
Population density in counties with reported TB419069 / county-population-density Evaluation notes Independent artifact and delivery checks passed. 56 tool calls	Computation	Pass	25m 39s	View task ↗
Map the distribution of lakes in relation to population centers in North America.422124 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	2m 6s	View task ↗
Electricity use in large-power-station countries429627 / spatial-analysis Evaluation notes Independent artifact and delivery checks passed. 52 tool calls	Computation	Pass	22m 32s	View task ↗
Generate a heatmap of power station density in regions with water scarcity435973 / geographic-heatmap Evaluation notes Selected feature membership differs from the independent answer. 78 tool calls	Computation	Fail	32m 17s	View task ↗
Calculate forest fragmentation indices for protected areas in Africa?436356 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 59s	View task ↗
Create a bivariate map showing the relationship between CO2 emissions per capita and greenhouse gases emissions per capita for 1990 and 2020 across all countries.438953 / map-series Evaluation notes Map-series boundaries differ from the original geography. 59 tool calls	Computation	Fail	17m 55s	View task ↗
Compare fertility rates between Eastern and Western African countries441352 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 51 tool calls	Computation	Pass	16m 16s	View task ↗
Compare healthcare infrastructure across developing economies.447179 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	2m 2s	View task ↗
Map the distribution of renewable water resources in arid regions.459898 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	2m 11s	View task ↗
Map total population distribution globally460208 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 50 tool calls	Computation	Pass	12m 12s	View task ↗
Make a heatmap of power station density in regions with high forest depletion470604 / geographic-heatmap Evaluation notes Final count or unknown-coverage count disagrees with the computed evidence. 74 tool calls	Computation	Partial	31m 44s	View task ↗
Map the distribution of tuberculosis cases across Massachusetts counties. 476053 / county-choropleth Evaluation notes Independent artifact and delivery checks passed. 48 tool calls	Computation	Pass	18m 37s	View task ↗
Show total labor force distribution across the world476358 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 58 tool calls	Computation	Pass	13m 45s	View task ↗
Find municipalities in Brazil where population density is above national average and are within 20km of both rivers and railways476928 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 40s	View task ↗
Map digital literacy rates in Southeast Asian nations.478472 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 4 tool calls	Limitation control	Pass	2m 24s	View task ↗
Make contour lines of population density in flood-affected areas of Bangladesh.480358 / contour-map Evaluation notes Independent artifact and delivery checks passed. 63 tool calls	Computation	Pass	24m 24s	View task ↗
What is a theodolite?483163 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 8 tool calls	Limitation control	Pass	3m 17s	View task ↗
Visualize infant mortality rates in South Asian countries.483194 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 3 tool calls	Limitation control	Pass	1m 57s	View task ↗
Generate a heatmap of mineral extraction facilities in regions with high water withdrawals in Africa486122 / geographic-heatmap Evaluation notes 40-minute task deadline exceeded.; No accepted final answer was published. 72 tool calls	Computation	Timeout	40m 0s	View task ↗
Map drought resilience indicators in semi-arid countries.487632 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	2m 32s	View task ↗
Create a heatmap showing population density near earthquake epicenters488600 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	2m 25s	View task ↗
Show urban green space percentage in European cities' countries.488696 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	2m 10s	View task ↗
Show the distribution of water stress in Mediterranean coastal areas?498483 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 55 tool calls	Computation	Pass	19m 0s	View task ↗
How does fertility rate vary across French-speaking African countries?501368 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 51 tool calls	Computation	Pass	14m 50s	View task ↗
Generate contour lines of population density near railways in upper-middle-income countries518135 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 3 tool calls	Limitation control	Pass	2m 22s	View task ↗
Show me map of electricity consumption in the World523121 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 49 tool calls	Computation	Pass	14m 30s	View task ↗
Compare water usage between Upper and Lower Nile countries526765 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 44 tool calls	Computation	Pass	14m 51s	View task ↗
Create contour lines showing population distribution in biodiversity hotspots526880 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 40s	View task ↗
Which African countries have the most efficient distribution of power stations relative to population?529248 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 5 tool calls	Limitation control	Pass	2m 1s	View task ↗
Map agricultural GDP contribution by region530802 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 52 tool calls	Computation	Pass	15m 57s	View task ↗
Display water stress levels in Middle Eastern and North African countries535461 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 47 tool calls	Computation	Pass	14m 10s	View task ↗
Compare population density between flood-affected and non-flood-affected areas in Bangladesh during August 2018536420 / observed-flood-population Evaluation notes Independent artifact and delivery checks passed. 63 tool calls	Computation	Pass	22m 46s	View task ↗
Map the distribution of healthcare expenditure as percentage of GDP in Western Europe.537597 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 20s	View task ↗
Calculate the average freshwater withdrawal in regions with significant mining activities in Africa538378 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 57s	View task ↗
Map marine protected areas in coastal nations.544781 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 4s	View task ↗
Calculate the percentage of Angola's population living in areas with density above 1000 people per square kilometer551060 / population-density-share Evaluation notes Independent artifact and delivery checks passed. 60 tool calls	Computation	Pass	15m 37s	View task ↗
Create a map showing total labor force distribution across countries553414 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 61 tool calls	Computation	Pass	15m 15s	View task ↗
Generate contour lines from snow cover data for 2023-2024 season and compare with railway stations locations in the northern states554817 / contour-map Evaluation notes Independent artifact and delivery checks passed. 45 tool calls	Computation	Pass	15m 34s	View task ↗
Display energy security metrics across Central Asian countries.558813 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 44s	View task ↗
Create a heatmap of earthquakes magnitude in Indonesia, Malaysia and Phillippines565545 / geographic-heatmap Evaluation notes Approval policy: injected feature data is not permitted.; No accepted final answer was published. 64 tool calls	Computation	Fail	22m 44s	View task ↗
Generate a heatmap of earthquake occurrences in regions with disaster preparedness572164 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 20s	View task ↗
Compare the agricultural value added between Brazilian states with and without railway access577156 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	2m 3s	View task ↗
Historical population near large wildfire incident locations586288 / raster-proximity-count Evaluation notes Independent artifact and delivery checks passed. 51 tool calls	Computation	Pass	26m 8s	View task ↗
Map water usage patterns in Middle Eastern countries595832 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 49 tool calls	Computation	Pass	15m 55s	View task ↗
Compare forest coverage between Amazon basin countries596430 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 57 tool calls	Computation	Pass	14m 44s	View task ↗
Make a heatmap of railway stations density in Brazil,596713 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 35s	View task ↗
Show agricultural productivity across Mediterranean countries598665 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 40 tool calls	Computation	Pass	12m 57s	View task ↗
Map freshwater withdrawals as percentage of resources602318 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 56 tool calls	Computation	Pass	24m 1s	View task ↗
Make a heatmap of seaport density in regions with high freshwater withdrawals in Latin America605895 / geographic-heatmap Evaluation notes Independent artifact and delivery checks passed. 73 tool calls	Computation	Pass	35m 59s	View task ↗
Show sustainable agriculture adoption in African countries.606612 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 43s	View task ↗
How many people live within 1 km from a railway in Bangladesh618282 / population-distance-share Evaluation notes Independent artifact and delivery checks passed. 49 tool calls	Computation	Pass	20m 6s	View task ↗
Create contour lines of population density along major rivers in Brazil620811 / contour-map Evaluation notes Independent artifact and delivery checks passed. 74 tool calls	Computation	Pass	32m 3s	View task ↗
Make a heatmap of seaport density along the Pacific coast of South America621471 / geographic-heatmap Evaluation notes Heatmap units do not match the declared weights per cell. 46 tool calls	Computation	Fail	19m 16s	View task ↗
Make a map visualizing relationship between fertility rate and net migration (normalized by total population) across the World.631643 / bivariate-map Evaluation notes Artifact measurement values differ from the frozen source calculation. 65 tool calls	Computation	Fail	23m 20s	View task ↗
Map total freshwater withdrawals as percentage of internal resources by country634036 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 45 tool calls	Computation	Pass	15m 46s	View task ↗
Counties touching the Mississippi River645898 / water-feature-selection Evaluation notes Independent artifact and delivery checks passed. 60 tool calls	Computation	Pass	17m 53s	View task ↗
Show public education spending across Scandinavian nations.646908 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 6 tool calls	Limitation control	Pass	1m 57s	View task ↗
Generate contour lines from Peru's population data and overlap with provincial borders and rivers in the country.655059 / contour-map Evaluation notes Contour map is missing a required context layer. 55 tool calls	Computation	Fail	16m 59s	View task ↗
Make a map to explore the relationship between forest area (% of land area) and annual freshwater withdrawals (% of internal resources) in 2015 globally.663841 / bivariate-map Evaluation notes Independent artifact and delivery checks passed. 56 tool calls	Computation	Pass	20m 22s	View task ↗
Compare environmental protection spending across developed nations.663887 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 30s	View task ↗
Generate a heatmap of population density along African railway corridors,664340 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 57s	View task ↗
Show the patterns of foreign direct investment across Asian economies.673093 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 3 tool calls	Limitation control	Pass	1m 48s	View task ↗
What percentage of Bangladesh's population lives within 5 km of a railway line?673274 / population-distance-share Evaluation notes Independent artifact and delivery checks passed. 51 tool calls	Computation	Pass	18m 8s	View task ↗
Show the distribution of renewable energy production by country.677223 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 3 tool calls	Limitation control	Pass	1m 34s	View task ↗
Create heatmap of TB cases in states with available data.695398 / county-heatmap Evaluation notes Independent artifact and delivery checks passed. 64 tool calls	Computation	Pass	23m 42s	View task ↗
Display net migration patterns on a world map702855 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 89 tool calls	Computation	Pass	24m 55s	View task ↗
Make a series of maps with share of forest covered areas in countries of South Asia in 2005, 2015 and the last available year.702872 / map-series Evaluation notes Independent artifact and delivery checks passed. 65 tool calls	Computation	Pass	18m 27s	View task ↗
How many mineral extraction facilities in Africa are located in countries with rapid urbanization710732 / country-change-selection Evaluation notes Independent artifact and delivery checks passed. 64 tool calls	Computation	Pass	26m 54s	View task ↗
Map absolute forest area distribution globally714649 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 58 tool calls	Computation	Pass	22m 50s	View task ↗
Create a map of global demographic patterns using fertility rates718627 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 61 tool calls	Computation	Pass	23m 20s	View task ↗
Create a heatmap of seaport density in regions with maritime development plans.719909 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 4 tool calls	Limitation control	Pass	2m 34s	View task ↗
Show electric consumption in G7 and BRICS nations720013 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 48 tool calls	Computation	Pass	16m 18s	View task ↗
Map total fertility rates across different regions720409 / country-choropleth Evaluation notes Final answer lacks one unambiguous requested machine-readable result. 73 tool calls	Computation	Fail	29m 28s	View task ↗
Make a heatmap of population density in areas with active wildfires in USA727667 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 4 tool calls	Limitation control	Pass	2m 6s	View task ↗
Which Eastern African countries had the highest net migration rates in 2019?734213 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 71 tool calls	Computation	Pass	23m 6s	View task ↗
Show total freshwater withdrawals by country737732 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 50 tool calls	Computation	Pass	15m 57s	View task ↗
Latin American seaports near supplied South American rivers741001 / point-line-proximity Evaluation notes Approval policy: explicit reduced or unavailable work is not preauthorized.; No accepted final answer was published. 36 tool calls	Computation	Fail	10m 24s	View task ↗
Show rural population percentages worldwide748943 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 64 tool calls	Computation	Pass	21m 37s	View task ↗
Make a World map of electricity consumption per capita by country761363 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 53 tool calls	Computation	Pass	18m 32s	View task ↗
Create contour lines from Chile's population density data in relation to rivers763412 / contour-map Evaluation notes Contour legend levels, distinguishable colors or units are incorrect. 94 tool calls	Computation	Fail	34m 7s	View task ↗
Generate a heatmap of population density in economic clusters765686 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 3 tool calls	Limitation control	Pass	2m 14s	View task ↗
Make contour lines showing population distribution in flood-prone areas of South Asia768309 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 6 tool calls	Limitation control	Pass	1m 36s	View task ↗
Visualize deforestation rates within 100km of major African highways?772258 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 4 tool calls	Limitation control	Pass	1m 52s	View task ↗
Make a heatmap final size of fire incidents in USA.782046 / geographic-heatmap Evaluation notes Required final-acreage observations are absent; incident sizes cannot substitute.	Computation	Setup blocked	Not recorded	View task ↗
Calculate the percentage of Angola's population living within 20km of mineral extraction facilities785163 / population-distance-share Evaluation notes Independent artifact and delivery checks passed. 64 tool calls	Computation	Pass	22m 4s	View task ↗
Display the variation in public education spending across Caribbean nations.792682 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 4 tool calls	Limitation control	Pass	1m 38s	View task ↗
Which African countries have the highest concentration of power stations in populated areas?794962 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 6 tool calls	Limitation control	Pass	2m 5s	View task ↗
Visualize fertility rates across Sub-Saharan Africa797835 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 41 tool calls	Computation	Pass	13m 43s	View task ↗
Water withdrawals in power-station countries806525 / spatial-analysis Evaluation notes Independent artifact and delivery checks passed. 49 tool calls	Computation	Pass	15m 0s	View task ↗
What is the total length of railways in Brazilian states with GDP per capita above national average?811107 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 4 tool calls	Limitation control	Pass	1m 36s	View task ↗
What percentage of Chile's population lives in areas with population density over 1000 people per square km?819657 / population-density-share Evaluation notes Independent artifact and delivery checks passed. 37 tool calls	Computation	Pass	14m 36s	View task ↗
Create a heatmap of earthquake occurrences in Pacific US states822439 / geographic-heatmap Evaluation notes Independent artifact and delivery checks passed. 46 tool calls	Computation	Pass	13m 54s	View task ↗
Show agricultural GDP contribution in Nordic countries826615 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 39 tool calls	Computation	Pass	13m 31s	View task ↗
Generate a heatmap of power station density in regions with renewable energy focus833409 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 32s	View task ↗
Make contour lines showing population distribution in economic corridors of South America833518 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 4 tool calls	Limitation control	Pass	2m 9s	View task ↗
Map water resource utilization globally833697 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 48 tool calls	Computation	Pass	14m 16s	View task ↗
Create a map of global fertility rates837688 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 36 tool calls	Computation	Pass	14m 13s	View task ↗
Visualize at one map rural population percentage (calculated from total and rural population) and agriculture value added as % of GDP for 2019.854050 / bivariate-map Evaluation notes Independent artifact and delivery checks passed. 62 tool calls	Computation	Pass	20m 25s	View task ↗
Display maternal mortality rates in West African countries.854925 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 45s	View task ↗
Visualize regional economic patterns using GDP per capita861416 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 54 tool calls	Computation	Pass	13m 44s	View task ↗
Compare healthcare expenditure across Middle Eastern countries.868995 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 53s	View task ↗
Snow-exposed Amtrak stations by state883928 / spatial-analysis Evaluation notes Independent artifact and delivery checks passed. 40 tool calls	Computation	Pass	14m 35s	View task ↗
Show agricultural GDP contribution in Himalayan nations894468 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 62 tool calls	Computation	Pass	15m 4s	View task ↗
Map the distribution of agricultural productivity in sub-Saharan Africa898861 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 41 tool calls	Computation	Pass	13m 50s	View task ↗
Compare GDP per capita between OECD and non-OECD high-income countries901809 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 41 tool calls	Computation	Pass	13m 20s	View task ↗
How does forest coverage vary across Amazon Treaty countries?903748 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 43 tool calls	Computation	Pass	12m 36s	View task ↗
Visualize total population distribution by country908870 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 59 tool calls	Computation	Pass	13m 26s	View task ↗
Generate a heatmap of power station density in regions with energy transition plans911221 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 49s	View task ↗
Generate a heatmap of power station density in African regions with high rural population911650 / geographic-heatmap Evaluation notes Final count or unknown-coverage count disagrees with the computed evidence. 97 tool calls	Computation	Partial	33m 49s	View task ↗
Show rates of deforestation over the last decade913900 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 42 tool calls	Computation	Pass	12m 52s	View task ↗
Visualize agricultural contribution to GDP worldwide914148 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 61 tool calls	Computation	Pass	14m 25s	View task ↗
Show rural population distribution across countries917242 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 53 tool calls	Computation	Pass	16m 31s	View task ↗
GDP in earthquake-affected countries918547 / spatial-analysis Evaluation notes Independent artifact and delivery checks passed. 46 tool calls	Computation	Pass	16m 14s	View task ↗
Create a heatmap of earthquakes in South Asia921361 / geographic-heatmap Evaluation notes Independent artifact and delivery checks passed. 50 tool calls	Computation	Pass	20m 3s	View task ↗
Map air pollution levels in East Asian countries.931200 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	1m 43s	View task ↗
How many African power stations are located in countries with significant forest loss932053 / country-change-selection Evaluation notes Independent artifact and delivery checks passed. 57 tool calls	Computation	Pass	20m 15s	View task ↗
Visualize the urban population growth in African capital cities over the last decade.938589 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 4 tool calls	Limitation control	Pass	1m 52s	View task ↗
Show population distribution in Caucasus region945038 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 57 tool calls	Computation	Pass	12m 26s	View task ↗
Map population density in Southern Asian nations946748 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 40 tool calls	Computation	Pass	15m 1s	View task ↗
Map food insecurity levels by administrative districts in Southeast Asia?954523 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	2m 0s	View task ↗
Counties shared by historical earthquake and wildfire snapshots955741 / county-event-intersection Evaluation notes Independent artifact and delivery checks passed. 48 tool calls	Computation	Pass	15m 33s	View task ↗
Map the relationship between population density and forest coverage in Asian countries.958938 / bivariate-map Evaluation notes Independent artifact and delivery checks passed. 58 tool calls	Computation	Pass	23m 26s	View task ↗
Create a world map showing forest area as percentage of land area by country960622 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 48 tool calls	Computation	Pass	12m 22s	View task ↗
Make a heatmap of population density in areas with high earthquake frequency965208 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 6 tool calls	Limitation control	Pass	2m 57s	View task ↗
Map rural population distribution in Central Asian nations965287 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 45 tool calls	Computation	Pass	12m 52s	View task ↗
Generate a heatmap of earthquake occurrences near railway lines in Canada968087 / geographic-heatmap Evaluation notes Independent artifact and delivery checks passed. 64 tool calls	Computation	Pass	16m 34s	View task ↗
Visualize agricultural value added as percentage of GDP worldwide971959 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 50 tool calls	Computation	Pass	12m 29s	View task ↗
Fertility comparison by income group977524 / spatial-analysis Evaluation notes Selected feature membership differs from the independent answer. 44 tool calls	Computation	Fail	14m 27s	View task ↗
Create population density contours around mineral extraction sites in low-income African countries979234 / qualitative-control Evaluation notes Operator-reviewed frozen qualitative criteria satisfied; not a numerical GIS result. 2 tool calls	Limitation control	Pass	2m 26s	View task ↗
Compare the freshwater withdrawal between African countries with and without significant railway networks.984439 / bivariate-map Evaluation notes Independent artifact and delivery checks passed. 45 tool calls	Computation	Pass	13m 50s	View task ↗
Map population distribution across English-speaking Caribbean islands986773 / country-choropleth Evaluation notes Independent artifact and delivery checks passed. 48 tool calls	Computation	Pass	12m 58s	View task ↗
Create contour lines of snow accumulation near major water bodies996706 / contour-map Evaluation notes Independent artifact and delivery checks passed. 53 tool calls	Computation	Pass	15m 14s	View task ↗

How the evaluation works

System-level evaluation

Our agent uses a wide variety of workflow, processor, catalog, and map tools to discover inputs, run processors, inspect artifacts and deliver results. We test the generated outputs against concrete expected values, feature selections and geometries calculated separately from the agent’s workflow. This evaluates the complete system, rather than isolating the underlying model.

01 / Inputs

Task data and configuration

Each task uses a separate workspace and checksum-verified source data. We record the code and model configuration; expected answers are withheld from the agent.

02 / Execution

Application runs and recorded outputs

We submit the question through the application API and record tool calls, workflows, outputs and the final answer, with a fixed deadline and no follow-up coaching.

03 / Grading

Artifact and answer checks

We compare generated data against reference results, verify that requested map layers were created, and check that the answer matches the data. Qualitative controls use explicit, AI-assisted and unblinded operator review against frozen criteria—not an external audit.

Evaluation date: 2026-10-07T10:04:59.842715+00:00
Model: gpt-5.4
Reasoning effort: Extra High
Application revision: d50497ea9084f29c91b05406688f87035df98d8a
Benchmark revision: 0edf6109ea5dca3e70fdfe081d39c9ccd0f832fb
Attempt policy: First valid primary attempt; verified infrastructure replacements only. Supplemental timeouts are separate.
Time limit per task: 2400 seconds
Tool-call limit: No separate tool-call cap; wall-clock limit applies
Grader revision: ebb3f5a7ba4114bff20137d5c9a92ce92f6bd56b2f8002ab694e8c32c15536a7

Limitations and caveats

Different tools, budgets and independent artifact grading prevent direct comparison with published GeoBenchX scores.
Dollar costs are unavailable. Qualitative judgments use AI-assisted operator review and are not blinded or an external audit.

We describe the original GeoBenchX evaluation for context; differences in models, tools, execution budgets and grading mean the scores are not directly comparable.

Comparison factor	GeoBenchX publication	This evaluation
Model	Published GPT-4.1, GPT-4o, o4-mini and other model configurations	gpt-5.4 / Extra High
Agent and tools	ReAct agent with 23 geospatial tools	Terra with Blue’s production tools
Grading	LLM-judge scoring against reference solutions	Artifact checks and explicit task-fulfillment criteria
Execution budget	Up to 25 agent iterations per task	2400 seconds per task; see protocol

Resources

Inspect

Download results CSV ↓Release manifest JSON ↓

Introducing Terra

The story behind Terra, how it works, and a step-by-step look at the evaluation.

Read the blog post ↗

Results & evaluation code

Versioned results, configuration, evaluation notes and scripts for checking the evidence.

GitHub repository ↗

Task evidence

Recorded outputs, observable agent activity and permitted artifacts supporting the reported outcomes.

Browse task evidence ↑

Original benchmark

GeoBenchX by Varvara Krechetova and Denis Kochedykov. See the original questions, methodology and attribution.

Read GeoBenchX ↗

Agent architecture

Analysis request

Agent harness

Catalog datasets & map layers

Vertical-slice evaluation

Trial workflow

Processor execution

Artifact inspection

Plan revision

Workflow execution

Saved workflow

Processor execution

Result assessment

Workflow revision

Simple Workflow Example

Results at a glance

GeoBenchX benchmark results

Explore the tasks

How the evaluation works

Task data and configuration

Application runs and recorded outputs

Artifact and answer checks

Limitations and caveats

Resources

Introducing Terra

Results & evaluation code

Task evidence

Original benchmark