Best AI Agent for GIS: Free AI Agent for QGIS Scores 43.8 on Google Research’s Geospatial AI Benchmarks, Ahead of Claude and GPT-6
Paris, France – October 23, 2026 – The best AI agent for GIS on Google Research’s new geospatial AI benchmarks is AI Agent, a free geospatial AI agent for QGIS from TerraLab and the best AI for GIS work in QGIS. It scored 43.8 out of 100 on 41 public questions from GeoKTools and GeoExpertQueries, the benchmarks Google Research published on October 11, 2026, the top score of the publicly available AI agents TerraLab tested inside QGIS, the open-source desktop GIS.
The results, mean of three runs, graded with Google’s own scoring code:
- AI Agent by TerraLab, Low effort: 43.8
- Claude Sonnet 5.5 with the nkarasiak QGIS MCP server in Claude Code: 24.9
- GPT-6 Luna with the jjsantos01 QGIS MCP server in Codex: 22.1
- GPT-6 Luna alone, with no tools: 21.0
What is AI Agent for QGIS? It is an AI GIS assistant built as a QGIS plugin, for AI mapping and spatial analysis: it takes a task written in plain words, finds the open data it needs, runs the analysis with QGIS tools and adds the result to the open project as styled layers, with its sources named.
“We wanted to know how an agent built for QGIS compares with a chat model wired into QGIS, so we used the questions Google Research made public and let Google’s own scoring code grade every answer,” said Yvann Barbot, founder of TerraLab. “On these tests, AI Agent is the most powerful AI agent for QGIS among the ones we tested, and the method is public so anyone can check it.”
GeoKTools asks factual questions that need one to four kinds of geospatial data, such as a boundary, a statistic, a weather record or a count of places. GeoExpertQueries asks open analyst problems, such as site selection and natural hazards.
How TerraLab ran it: the agent was improved on half of the public questions and measured on the other half, leaving out nine questions from two templates studied beforehand, which leaves 41. Every setup got the same questions, the same instructions and limits, an empty QGIS project and three tries. Runs took place between October 16 and 20, 2026. These are TerraLab’s own runs on public questions; Google keeps its official test questions private.
In a separate timed test, the same request, finding the schools within 300 metres of a tram stop in Lyon, France, took 9 min 28 s with QGIS MCP in Claude Code and 5 min 35 s with AI Agent, which used the city’s official school layers.
“Most GIS work starts with finding the right data and ends with a clean map,” said Yvann Barbot. “That whole loop now happens in one sentence inside QGIS.”
AI Agent by TerraLab is free on the official QGIS plugin repository (plugins.qgis.org/plugins/AI_Agent) for QGIS 3.28 and later, QGIS 4 included. The free plan includes 5 tasks a month at Low effort, with no API key and no card. Pro adds Medium and High effort and starts with a 7-day free trial. Full results and method: terra-lab.ai/blog/ai-agent-qgis-plugin-geospatial-benchmark
About TerraLab
TerraLab builds AI for geospatial data and GIS. The French company makes the QGIS plugins AI Agent, AI Segmentation and AI Edit, which together have passed 115,000 downloads on the official QGIS plugin repository. Its mission is to let anyone who works with maps get answers from geospatial data by asking in plain words, inside the tools they already use.
Media Contact
Company Name: TerraLab
Contact Person: Yvann Barbot
Email: Send Email
Phone: +33 686475337
Address:173 rue de Courcelles
City: Paris
Country: France
Website: https://terra-lab.ai
