STEP, JT and 3D PDF: choosing a neutral CAD format

Neutral CAD formats get discussed as though they were rivals, with the question framed as which one is best. That framing produces bad decisions, because the three that matter — STEP, JT and 3D PDF — were built for different jobs and are each poor substitutes for the others.

A more useful question: what will the recipient do with this file?

What the recipient does determines the format

Three broad answers, and they map cleanly onto the three formats.

They will build on the geometry. A supplier importing your model to design a fixture, run simulation, or program a machine tool needs geometry that is exact and modifiable. That means boundary representation — the mathematical surface definition — not a mesh approximating it.

They will look at the geometry. A design review, a digital mock-up, a clash check across a full vehicle or aircraft. Precision matters less than being able to open something with a hundred thousand components without waiting ten minutes.

They will read the geometry. Someone in procurement, quality or the shop floor who needs to see the part and its dimensions, and who does not have CAD and is not going to install any.

STEP: the exchange and archival format

STEP is ISO 10303, standardised since the mid-1990s, and it is the default answer for CAD-to-CAD exchange. It carries exact B-Rep geometry, assemblies, and metadata such as materials and tolerances.

The part worth knowing is the application protocol. STEP is not one format but a family, and the AP determines what survives the trip. AP203 and AP214 are the long-standing mechanical design protocols. AP242 is the one to ask for when model-based definition matters, because it carries product manufacturing information — and, critically, carries it semantically.

STEP is also the format for long-term archival. Where retention obligations run to decades, as in aerospace and defence, AP242 is what the LOTAR archival standards are built around. Native CAD formats are not archival formats; they depend on a vendor and a version that will not necessarily exist in thirty years.

JT: the visualisation format

JT is ISO 14306, originally developed by Siemens and now maintained alongside the ProSTEP iViP/VDA consortium. It was the first 3D visualisation format to reach international standard status.

Its structure explains what it is for. Every JT file contains a mandatory tessellation at multiple levels of detail — triangle meshes at varying resolutions — which is what allows a viewer to load an enormous assembly quickly and refine detail as you zoom. Compression is aggressive, commonly cited in the range of 10:1 to 50:1.

Exact B-Rep in JT is optional. It can be present as Parasolid XT (the industry-recommended variant) or STEP B-Rep, or it can be absent entirely.

That optionality is the trap. A JT file without embedded B-Rep supports neither precise measurement, nor simulation, nor redesign. It looks like a model and behaves like a picture. If you are sending JT to a supplier who will do engineering work with it, you need to know which variant you are producing, and so do they.

3D PDF: the distribution format

3D PDF embeds 3D data — usually PRC, which can carry exact B-Rep as well as tessellation — inside a PDF that opens in a reader the recipient already has.

Its advantage is entirely about reach. No CAD licence, no viewer installation, no training. For a quality report, a service instruction, or a drawing package going to someone outside engineering, that matters more than any technical property.

Its limitation is the mirror image: CAD systems do not generally import 3D PDF. It is a terminal format. Data goes in and is read by humans; it does not come back out into a design workflow. Choosing it for supplier exchange means choosing to send data that cannot be used to build anything.

The PMI trap

If you take one thing from this, take this one.

Model-based definition moves tolerances, annotations and specifications off the drawing and into the model. But PMI comes in two forms, and they are not interchangeable:

Semantic PMI retains a programmatic link to the geometry it applies to. A tolerance knows which face it constrains. Software downstream can read it, act on it, drive inspection from it.
Graphical PMI is the appearance of that annotation without the link. It displays correctly and means nothing to a machine.

Cross-system conversion is where semantic PMI degrades to graphical. The annotation appears, the file looks right, a human reviewing it sees no problem — and every automated downstream use is silently broken. You do not discover this at conversion time. You discover it when someone tries to generate an inspection program and finds there is nothing to generate from.

This is why AP242 specifically, rather than STEP generally, is the answer for MBD delivery. And it is why “we exchange in STEP” is an incomplete statement: the protocol and the PMI mode are the parts that determine whether MBD survives.

A short decision guide

The recipient will	Use	Watch for
Design, simulate or machine from it	STEP AP242	Confirm PMI is semantic, not graphical
Review or clash-check a large assembly	JT	Know whether B-Rep is embedded
Read it without CAD	3D PDF	It is one-way; nothing comes back
Archive it for decades	STEP AP242	Native CAD formats are not archival

What to check before standardising

Test with your actual data, not a sample part. Translation quality varies enormously with geometry. A format that handles a bracket cleanly may mangle a surfaced body.

Verify PMI semantically, not visually. Open the converted file and confirm a tolerance still resolves to a face. If your only check is that it looks right, you are not checking the thing that breaks.

Ask suppliers what they can actually consume. The format you standardise on is worthless if half your supply chain has to convert it again on receipt, since each conversion is another opportunity to lose the semantic layer.

Separate the exchange format from the archive format. They have different requirements and there is no obligation to use one format for both.