Text-to-CAD in 2026: Every Tool Tested and What Actually Works for Engineers

We tested every text-to-CAD tool in 2026. Most generate mesh output, not parametric CAD. Here's what works and what doesn't for real engineering.

May 12, 2026

⏱

10 min

Michelle Ben-David

Product Specialist, Leo AI

Mechanical Engineer, B.Sc. · Ex-Officer, Elite Tech Unit · Aerospace & Defence · Medical Devices

Michelle Ben-David is a mechanical engineer and Technion graduate. She served in an IDF elite technology and intelligence unit, where she developed multidisciplinary systems integrating mechanics, electronics, and advanced algorithms. Her engineering background spans robotics, medical devices, and automotive systems.

Introoduction

Zoo.dev (Formerly KittyCAD): The Closest Thing to Real Text-to-CAD

Autodesk Fusion AI: Generative Design with Guardrails

SolidWorks AURA: Not What the Headlines Suggested

Other Players: Where the Rest of the Field Stands

The Problem Everyone Misses: You Probably Already Have the Part

Bottom Line

FAQ

Glossary

BOTTOM LINE

Text-to-CAD tools in 2026 are improving but still mostly produce mesh geometry, not production-ready parametric CAD. SolidWorks AURA turned out to be a documentation chatbot, not the generative design engine the demos promised. For most engineering teams, the bigger win is not generating new geometry from text - it is finding the existing validated parts already sitting in your PDM vault. Leo AI's patented geometry-aware search does exactly that, across every major PLM platform.

The pitch sounds incredible. Type "flanged bearing housing with M8 mounting holes" into a text box, and a fully parametric CAD model appears on your screen. Ready for manufacturing. Ready for your assembly. That is the promise the text-to-CAD market has been selling for the past two years.

The reality is quite different. Most of these tools generate mesh geometry - STL or OBJ files that look nice in a viewport but fall apart the moment you try to edit them, tolerance them, or send them to a machinist. Some tools are vaporware. Others are documentation chatbots dressed up as generative design engines. And a handful do genuinely useful things, just not what the marketing suggests.

I spent weeks testing every major text-to-CAD tool available in 2026. Here is what actually works, what does not, and what the whole conversation gets wrong about how engineers should be using AI for part creation.

Zoo.dev is the tool most people think of when they hear "text-to-CAD." Their approach generates 3D geometry from text prompts using a purpose-built model, and they deserve credit for tackling a genuinely hard problem head-on.

The output quality varies. Simple shapes - brackets, plates, basic enclosures - come out reasonably well. More complex geometry gets unpredictable. The bigger issue is what format that geometry lands in. Most outputs are mesh-based, not parametric BREP solids. You get something that looks like a bracket, but you cannot grab a fillet radius and change it to 2mm. You cannot suppress a feature or drive a dimension from a design table.

For engineers doing conceptual exploration or generating quick visual mockups, Zoo.dev has real value. For anyone who needs to hand a file to manufacturing or drop it into a production assembly, the gap between mesh output and parametric CAD is still a dealbreaker.

IN PRACTICE

The part search capabilities are really in a league of their own - text to text, text to CAD, and CAD to CAD. It's really something you have to try for yourself to see.

"The part search capabilities are really in a league of their own - text to text, text to CAD, and CAD to CAD. It's really something you have to try for yourself to see." - erga k., Product Engineer

Autodesk has been layering AI capabilities into Fusion for a while. Their generative design tools let you define constraints - loads, boundary conditions, keep-out zones - and the system generates optimized geometry.

This is not really text-to-CAD in the way most people imagine it. You are not typing a sentence and getting a model. You are setting up an engineering optimization problem and letting the solver explore the design space. The results are often organic, topology-optimized shapes that look impressive but require significant cleanup before they are manufacturable.

Fusion's AI features work well within their intended scope: early-stage exploration where you want to understand what the physics allows. But calling it "text-to-CAD" stretches the definition. And the output still typically needs manual conversion to production-ready parametric geometry.

AURA got enormous attention when Dassault showed it at 3DExperience World 2025. The headline demo was stunning - type a text prompt, get a full assembly with sub-assemblies and part files generated automatically. That demo made it look like SolidWorks had cracked the text-to-CAD problem.

Here is what actually shipped: AURA is a documentation chatbot built on Mistral AI. It answers "how do I..." questions about SolidWorks operations. It searches SolidWorks help files, documentation, and community forums. It can retrieve results with source links and help with SolidWorks community content.

What it does not do: generate CAD geometry from text prompts. The assembly generation demo from 3DX World 2025 never shipped. AURA does not do generative part creation, does not do topology optimization (that is SolidWorks Simulation, a separate pre-existing product), and does not access your PDM vault data.

There is also the platform lock-in issue that rarely gets mentioned. AURA requires 3DExperience Connected. If you are running traditional desktop SolidWorks - which the majority of the roughly 700,000 professional license holders are - you cannot use AURA without purchasing new cloud licenses. That is a significant barrier that most reviews skip over.

As of mid-2026, AURA remains in beta with only its chatbot functionality available. It is useful for SolidWorks help questions. It is not a text-to-CAD tool.

Several other tools claim text-to-CAD capabilities in various forms. Most fall into a few categories.

Mesh generators produce STL or OBJ output from text descriptions. The geometry might look correct on screen, but you cannot edit features, drive dimensions, or generate proper engineering drawings from mesh data. For 3D printing prototypes, these work. For production engineering, they do not.

Script generators use LLMs to write OpenSCAD or CadQuery code from text descriptions. The output is parametric, which is good, but the geometry is limited to what you can express programmatically in those languages. Complex surfacing, lofts, or multi-body configurations are either impossible or extremely fragile.

API wrappers use AI to send commands to existing CAD platforms through their APIs. The results depend entirely on the quality of the underlying CAD engine, and the AI layer is essentially translating text into a sequence of modeling operations. These can work for simple parts but break down quickly with complexity.

None of these approaches produce the kind of production-ready, fully parametric, natively-editable CAD files that engineers actually need. The gap between "AI-generated geometry" and "file I can send to my machine shop" remains wide.

Here is the thing nobody talks about in the text-to-CAD conversation. Most of the time, an engineer does not need to generate a new part from scratch. They need to find an existing one.

Studies consistently show that 60-80% of new parts designed in engineering organizations are functionally similar to parts that already exist in the company vault. Engineers create duplicates because searching PDM systems is painful. Metadata is inconsistent. File naming conventions drift over time. The search tools built into most PLM platforms require you to know exact part numbers or filenames - and if you knew those, you would not be searching.

This is where the text-to-CAD conversation takes a wrong turn. The industry is pouring resources into generating new geometry from text, when the more impactful capability is finding existing proven geometry from text.

FAQ

Find Parts, Don't Reinvent Them

Search your entire vault with natural language

Leo AI searches your PDM vault using text descriptions or even uploaded CAD geometry. Find existing parametric parts in seconds instead of designing duplicates from scratch.

Schedule a Demo →