Text2CAD
Around 170K CAD models with multi-level prompts from abstract to expert, generating sketch-and-extrude sequences with an end-to-end transformer model.
Summary rewritten from the vendor site because the catalog version did not match it.
What it does and how it connects
Text2CAD is the first AI framework for generating text-to-parametric CAD models using designer-friendly instructions for all skill levels. It introduces a data annotation pipeline that generates text prompts from the DeepCAD dataset using Mistral and LLaVA-NeXT, covering abstract CAD descriptions through to detailed specifications.
The dataset contains around 170K models and around 660K text annotations, from abstract descriptions such as generating two concentric cylinders to detailed specifications that name circle centers, radii and extrusion depths. Within the framework, an end-to-end transformer-based auto-regressive network generates parametric CAD models from input texts as sketch-and-extrude sequences.
The paper evaluates the model on visual quality, parametric precision and geometrical accuracy, and reports that the framework shows great potential for AI-aided design applications.
Alternatives
Compare side by side- Text2CAD-Bench
A benchmark that scores text-to-CAD generation on 600 human-curated examples across four complexity levels, rather than a framework that generates CAD models from text prompts.
- DeepCAD (dataset)
A dataset of 178K sketch-and-extrude sequences parsed from Onshape, which Text2CAD's annotation pipeline builds on, rather than the text-to-CAD framework itself.
- CADBench (Doris et al.)
A unified 18,000-sample benchmark turning mesh, image or render inputs into STEP or mesh, rather than a text-to-CAD generation framework.
- MUSE
Grades text-to-CAD assemblies in a three-stage funnel ending in a VLM judge, rather than generating parametric CAD models from text with a transformer model.
- BenchCAD
Scores execution-verified CadQuery programs by deterministic voxel IoU across 106 industrial part families, rather than generating sketch-and-extrude sequences from text prompts.
Sources
Not verified
- License: no license is stated on the arXiv page read.
- A public repo: the abstract states source code and annotations will be publicly available, but no repository URL appears on the abstract page read.
- Pricing: none stated (research framework).
- Real user voices: no real linked post read for this page.
- Related Error Index entries: no matching entry verified.
- The directory record's 150K model count: the arXiv abstract states around 170K models, so the one-liner was rewritten to the abstract's figure.
Last verified November 6, 2026. Listings are informational. The evidence score measures public, verifiable evidence and is not an endorsement or a compatibility claim.