QueryCAD
111 questions over 18 CAD models testing CAD understanding: the LLM writes Python against a CAD interface to answer measurement and feature queries.
What it does and how it connects
QueryCAD is the first system designed for CAD question answering, extracting precise information from CAD models using natural language queries. The LLM writes Python against a CAD interface to answer measurement and feature queries, so answers are grounded in the model rather than guessed from a description.
It incorporates SegCAD, an open-vocabulary instance segmentation model developed by the authors to identify and select specific parts of the CAD model based on part descriptions. The paper also proposes a CAD question answering benchmark to evaluate QueryCAD and establish a foundation for future research.
QueryCAD is integrated within an automatic robot program synthesis framework, where the ability to process CAD models lets deep-learning solutions for robotics use the information those models contain.
Alternatives
Compare side by side- BenchCAD
Scores execution-verified CadQuery programs by deterministic voxel IoU across 106 industrial part families, rather than answering measurement and feature questions about CAD models.
- MUSE
Grades text-to-CAD assemblies in a three-stage funnel ending in a VLM judge, rather than answering natural-language questions about existing CAD models.
- Text2CAD-Bench
Tests text-to-parametric-CAD on 600 human-curated examples across four complexity levels, rather than a question-answering benchmark over CAD models.
- CADBench (Doris et al.)
A unified 18,000-sample benchmark turning mesh, image or render inputs into STEP or mesh, rather than a CAD question-answering benchmark.
Sources
Not verified
- License: no LICENSE file is in the repo and the GitHub API returns no license, so no license is stated.
- The 111-question count in the one-liner: directory record; the README confirms 18 CAD models but does not state a 111-question total.
- Pricing: none stated (research system).
- Real user voices: no real linked post read for this page.
- Related Error Index entries: no matching entry verified.
Last verified November 6, 2026. Listings are informational. The evidence score measures public, verifiable evidence and is not an endorsement or a compatibility claim.