Your SOLIDWORKS Assembly Is Slow Because of the Bolts. LEO in FD03 Tries to Say Which Ones

Large SOLIDWORKS assemblies get slow for mechanical reasons the software has had for years. Every fastener with a modeled helix thread is a pile of graphics triangles. Too much detail sits loaded at the top level. Rebuilds get expensive. Mates pile up. Imported junk stays in the tree. Opening and rebuilding can take tens of seconds to minutes.

A helix is a lot of faces. A hundred bolts with real threads is millions of triangles for geometry that, on the shop floor, is a callout on a drawing. Cosmetic threads exist for this reason. People still import hardware with real helices because Toolbox, a supplier STEP, or a habit put them there.

The old tools already exist: Performance Evaluation, Assembly Visualization, Large Design Review, SpeedPak, cosmetic threads. They work if you know which one to use and where to click. SpeedPak is the lightweight stand-in. Large Design Review is the "just look at it" mode. You still have to decide when.

The other half of the pain is the export ritual nobody wants to write down. Dump every part to STEP. Rename to a release scheme. Batch PDF. Purchasing wants the STEP. The vault wants a name. QA wants a PDF. None of that is a feature you click once unless someone wrote the macro. That is VBA, or a VAR, or a Saturday.

Two moves from LEO

Dassault's answer in 2026x is LEO, the virtual companion in SOLIDWORKS Design. It is supposed to do this in two moves, and they did not ship on the same day.

Move A is diagnosis. It actually landed in R2026x FD02 on 29 April 2026 as "Assembly Performance Evaluator (Beta) – get real-time advice on how to speed up assembly performance." You ask LEO something like "why is this assembly slow," "evaluate my assembly for performance," or "how do I optimize this assembly." The vendor claim, restated in the 5 August AI post, is that LEO looks at the active assembly (open time, rebuild, graphics load, structure) and returns a prioritized list instead of you hunting component by component:

Improve large assembly performance by asking LEO why an assembly is slow and how to optimize it, rather than manually inspecting every component and performance setting. LEO analyzes open time, rebuild time, graphics load, and structural complexity, then translates the results into a clear, prioritized breakdown of the issues having the greatest impact.

Move B is action, plus macros. That is FD03 (platform drop around 11 July, AI blog 5 August). LEO is supposed to act on that list: convert some modeled threads to cosmetic representations, recommend or create SpeedPak at the top level or a subassembly. Separately, it is supposed to generate a VBA macro from a sentence (renames, exports, properties) and open it in the editor for review.

Vendor language on threads and SpeedPak:

LEO can convert modeled threads on imported fasteners and other components into cosmetic thread representations across the entire assembly… LEO can also recommend and create the appropriate SpeedPak configurations at the top-level assembly or subassembly level based on assembly structure, component usage, and performance needs.

On macros:

Ask LEO to create a macro for tasks such as renaming files, exporting deliverables, applying company standards, or updating model properties, and LEO generates the macro and opens it directly in the editor for review.

No customer names and no times on that page.

What is actually public

As of 29 August 2026, no named end-user company has published a production case study of the diagnosis skill or the acting and macro skills. Coverage is vendor blogs, vendor demos, and reseller recaps.

The FD03 wiki is blunt: "All AI capabilities listed below are in Beta state as of July 11, 2026. Features are subject to being removed from Beta at any future release."

The numbers you will see in recaps are Dassault demos, not plant results.

A wiki example on an unlabeled assembly (no hardware listed, no customer named) says: "In this example, LEO reduced assembly open time from 52 seconds to 8 seconds, about 84% faster."
A YouTube Short (vendor marketing; the narrator says "our last case study" and names nobody) uses the prompt "evaluate my assembly for performance." Fasteners are "over 4.5 million triangles"; suppressing them is said to "reduce the rebuild time by over 10 seconds, which is about a 30 to 50% improvement." "all within two prompts." That is a what-if suppress-fasteners demo, not SpeedPak or cosmetic-thread conversion, and not a named customer.

Delivery Day on 30 July, with Sean O'Neill and Jesse Sprague of SOLIDWORKS staff, was filmed at Vatn Systems, a real AUV customer. Sophie Beitel, a Vatn mechanical engineer, was on camera for other FD03 features (Pattern Assist, Large Design Review, and the like). She was not asked about LEO diagnosis, SpeedPak-from-LEO, thread conversion, or generated macros. Vatn hosted the video. That is not a LEO performance or macro case study.

On camera, staff ran the prompt "how do I optimize this assembly." LEO inspects the active assembly and can offer SpeedPak because a subassembly hurts open time. Thread conversion, O'Neill said, suppresses a lot of modeled threads faster than a human, but cannot get all of them. It gets cut, sweep, and chamfered threads. "imported body" or "thread tool" threads: "it's not going to get those as of today." The worked macro example was export all parts in the active assembly as STEP, then open the VBA editor. You can save a .swp and pin it as a button. Sprague, on the macro skill: "I haven't used it myself yet, but I'm really looking forward to using this."

Resellers recapped. They did not publish runs.

TECHNIA's Michael Hopley-Ash, CSWE, on macros: "While I don't expect results to be perfect straight away, there's a fun iterative element… removing the initial barrier of having to learn to code…" No sample prompt, no time saved.

GoEngineer flagged the thread caveat ("Not every thread type is currently supported") and the obvious macro rule: "Review the macro and test it on non-production data before using it more broadly."

ChampionXperience on 12 August is a paraphrase of the official post. They invented a DXF sheet-metal prompt that is not on the official page. Treat that as their example sentence, not as something anyone ran.

Reddit, forum.solidworks.com, and Eng-Tips had nothing found for LEO FD03 diagnosis or macros as of 29 August 2026.

The catch

Thread conversion misses some types. O'Neill said so on camera. GoEngineer said so in print.

Generated macros need review. They should be tested off production. Sprague had not used the macro skill himself on the launch demo.

Tokens exist. A 13 August vendor explainer lists 2,100 intro tokens and lists assembly performance analysis as token-consuming. Version notes conflict across Dassault pages (SP3.0 vs SP3.2). Don't treat one of those strings as gospel.

Beta means it can still be pulled. The wiki said that on 11 July.

The Monday experiment

Here's the experiment if you already have 2026x FD03 with LEO. A recipe, not a completed trial.

Duplicate a slow assembly. Not the vault copy. Open the duplicate. Ask why it's slow, or "evaluate my assembly for performance," or "how do I optimize this assembly." Read the list against Performance Evaluation yourself. Do not let it write SpeedPak or convert threads on the production file until you've checked the list.

For macros, ask it to export STEP of the parts in the active assembly. That is the only on-screen example Dassault showed. Read the VBA. Run it on a sandbox folder, not PDM.

If the list and Performance Evaluation disagree, believe the stopwatch and the FeatureManager, not the chat pane.

What it isn't

It isn't a new physics of assemblies. The triangles, rebuilds, and mates were already the problem. The old tools still exist. LEO is supposed to point at them, then click some of them for you.

It isn't a plant result. The 52-to-8-second wiki example and the 4.5-million-triangle Short are Dassault demos. Vatn hosted a video. They were not asked about these skills.

If you don't have FD03 and LEO, the Monday work is still Performance Evaluation, cosmetic threads, SpeedPak, and a VAR or a Saturday for the STEP dump.

Sources

Official 5 August 2026 AI post (diagnosis, threads, SpeedPak, macros): https://blogs.solidworks.com/products/solidworks/whats-new-in-solidworks-ai-r2026x-fd03/
FD02 design and modeling (29 April 2026; Assembly Performance Evaluator Beta): https://blogs.solidworks.com/products/solidworks/whats-new-in-solidworks-design-r2026x-fd02-design-and-modeling/
FD03 wiki (AI still Beta as of 11 July 2026): https://3dswym.3dexperience.3ds.com/wiki/solidworks-news-info/solidworks-design-r2026x-fd03_Xl878wGUQ2Ct7lXCWZVKxQ
Virtual Companions wiki (52s-to-8s vendor example): https://3dswym.3dexperience.3ds.com/wiki/solidworks-news-info/what-s-new-generative-experiences-virtual-companions_dUETx-VASWaynZGRLPgXrg
YouTube Short, vendor marketing (4.5M triangles / suppress-fasteners demo): https://www.youtube.com/shorts/86tNIJFTsYE
Delivery Day, 30 July 2026 (staff demo filmed at Vatn Systems): https://www.youtube.com/watch?v=gKUpSmMSqXs
TECHNIA recap, Michael Hopley-Ash: https://blog.technia.com/en/engineering/top-3-features-solidworks-r2026-fd03
GoEngineer, "AI in SOLIDWORKS": https://www.goengineer.com/blog/ai-in-solidworks
ChampionXperience, 12 August paraphrase: https://blog.championxperience.com/solidworks-ai-r2026x-fd03-leo-companion/
Tokens explainer, 13 August 2026: https://blogs.solidworks.com/products/solidworks/solidworks-ai-101-understanding-how-tokens-work/