alt.hn

5/22/2026 at 10:38:26 AM

Antigravity 2.0 Tops the OpenSCAD Architectural 3D LLM Benchmark

 https://modelrift.com/blog/openscad-llm-benchmark/

by jetter

5/22/2026 at 11:59:58 AM

Last weekend I bought my wife a bike off marketplace. It was in good condition but was missing one of the internal cable routing grommets. I gave Claude pictures of the pill-shaped hole by itself and with my digital calipers in the long and short directions.

Gave it a short prompt and it gave me an openscad model with everything parametrized. I printed with no changes in tpu and it was nearly perfect on the first try. Claude put in a 0.3mm subtraction in the x/y dimensions and I lowered it to 0.1 and it's perfect.

Much easier shape than ancient Roman architecture but still very cool how easy it was.

by jhot

5/22/2026 at 12:20:39 PM

Yeah, CAD has been my personal example of "oh the barrier to entry for this skill was high enough that I didn't do it and now I can be passably bad at it enough to get some simple things done"

I've had similar experiences with making simple functional parts off a 3d printer with OpenSCAD + LLMs. I'm very aware that the models are worse at it than say, generating react code, and I'm also the antithesis of a skilled pilot. It's still cool and has resulted in me starting to learn a new skill at a hobby level.

by simplyluke

5/22/2026 at 3:31:10 PM

It's like this with a lot of things now. For example, Nix's learning curve used to be a huge barrier to entry. Now with LLMs, I'm using nix-darwin and home-manager for dotfiles, package management, and have individual flakes in all of my projects for cryptographically reproducible builds!

by dempedempe

5/22/2026 at 4:28:08 PM

Nit: there’s nothing “cryptographic” about reproducible builds.

“Reproducible build” already usually implies bit-by-bit reproducibility.

by rlt

5/22/2026 at 8:16:23 PM

“The reproducibility is cryptographically verifiable with hashes“ would be the full sentence, but it’s a mouthful.

by illiac786

5/23/2026 at 2:40:25 AM

Build reproducibility checks usually use bitwise comparison, not hash comparison.

The Reproducible Builds project also wrote diffoscope, which goes quite far with helping identify where differences occur and how to fix them.

https://reproducible-builds.org/ https://diffoscope.org/ https://try.diffoscope.org/

by pabs3

5/23/2026 at 8:16:16 AM

Let’s say, for the positive case, hash comparison is significantly faster.

by illiac786

5/23/2026 at 8:38:54 AM

I feel like that is quite unlikely. Both the hash and bitwise comparisons read both files in both cases. In the not-equal case the hash reads the entirety of both files, so its slower than a start-to-end bitwise comparison, which exits at the first not-equal bit. In the equal case, both read the entirety of both files. Various other bitwise strategies can be faster than start-to-end, rdfind for example checks the start of the file first, then the end, then the rest of the file.

by pabs3

5/23/2026 at 11:56:56 AM

I think we’re not talking about the same scenario. I’m talking about the case where at least one hash has already been calculated.

by illiac786

5/22/2026 at 10:24:59 PM

yes, but it's still not cryptological, it's just verification using hashes.

by dekhn

5/22/2026 at 10:43:42 PM

The hash being cryptographically secure is significant. In contrast, you could use (for example) md5 to non-cryptographically verify that the full process matched.

by fc417fc802

5/23/2026 at 1:12:00 AM

Sorry, the point I was making is that this isn't cryptography- it's the properties of a cryptographic hash (hard to spoof) that are useful. I don't think any verified build program uses the hash to encrypt data at any point. If I'm wrong on this point, that's fine, but please include a link.

by dekhn

5/23/2026 at 1:19:08 AM

Sure, "verified in a cryptographically secure manner" is technically not equivalent to "cryptographically verified" but the response "it's not cryptographic" is rather ambiguous at best given that it is, in fact, a cryptographically secure manner of verification. The key observation here being that an algorithm or process being "cryptographically secure" does not mean that it is "cryptographic" in nature (ie implements or uses cryptography).

by fc417fc802

5/22/2026 at 5:56:41 PM

I meant with Nix you're comparing hashes. With Docker, you're using pinned versions

by dempedempe

5/22/2026 at 5:07:08 PM

i thought it mainly implied architectural/hardware compatibility and deterministic output

by bt1a

5/22/2026 at 11:16:10 PM

Nix mostly does not guarantee deterministic output. It rather guarantees deterministic inputs, and then sandboxes the system to inhibit the build from accessing the outside world.

Deterministic inputs do not always imply deterministic outputs.

by aidenn0

5/23/2026 at 2:41:25 AM

Indeed, the Reproducible Builds community is working on fixing non-deterministic build output https://reproducible-builds.org/

by pabs3

5/22/2026 at 5:08:54 PM

Nix is also great at work. You keep the server nix code in the same repo and OpenCode can just change and test server config.

by pimeys

5/22/2026 at 1:04:48 PM

Learning to make simple parts in onshape is pretty darn easy (and fun).

by 0x696C6961

5/22/2026 at 2:40:46 PM

Yeah. I teach this after school to 7th grade kids. Anyone can pick this up in a few hours.

by jeffbee

5/22/2026 at 3:18:14 PM

They taught us to make Legobricks with CAD when I was in 6th. Wish I retained more of that and that it would be more widely taught.

by chalupa-supreme

5/22/2026 at 5:40:59 PM

I am reasonably confident that access to solid modeling and additive fabrication is now more widespread than ever.

by jeffbee

5/23/2026 at 2:59:24 AM

I mean, like any other skill that has pretty much been my experience (though I tried fusion + openscad), but there is something about being able to ask a computer all the dumb noob questions that makes that first phase easier.

by simplyluke

5/23/2026 at 1:27:39 AM

same — LLMs turn skills i'd parked for years into 'just try it' territory, which is genuinely new.

by k-Whale

5/22/2026 at 4:15:21 PM

Claude does well if you can provide all dimensions. It fails at guessing though. The real magic is when you can provide one dimension or photograph with a ruler in it and the AI will figure the rest out. Right now, Claude anyways, is pretty bad at guessing.

by skinner927

5/22/2026 at 2:38:11 PM

I was recently trying to get models to generate a 3D fortune cookie. Claude in three.js and Gemini in openSCAD. Neither really got the concept or could get very close at all. It's a surprisingly complex shape I guess.

by jonah

5/22/2026 at 11:41:21 PM

Probably easier with Trellis 2 or Meshy.ai

by car

5/22/2026 at 5:41:08 PM

with the shape you probably want something thats good at bends/fabric

cause youd start with the flat shape, the set some contraints that certain edges are colinear

by 8note

5/23/2026 at 3:36:36 PM

Amazingly I tried the exact same thing! It also kept making meaningless corrections and was so confident that the blob it had produced was a fortune cookie that it started telling me I was wrong.

by idontwantthis

5/22/2026 at 12:14:18 PM

these small functional prints are exactly where OpenSCAD and LLM generation shines

by jetter

5/22/2026 at 12:40:58 PM

Does it optimize for no support?

by amelius

5/22/2026 at 1:15:31 PM

You optimize for no support when selecting print orientation (but for anything semi-cylindrical like described that would be the only sane orientation and the one slicer would choose when you smash the 'Auto Orientation' button).

by 05

5/23/2026 at 3:56:20 PM

Naive/uninformed question: does 3D printing produce any cleavage issues (like you’d get with a crystal) along the directions of printing, and does it ever make sense to change the orientation to help with that?

by bee_rider

5/23/2026 at 9:15:44 PM

Of course it does! Inter-layer adhesion is lower than the adhesion within the same layer, so it's often recommended to orient the print in a way that maximizes strength - especially for tensile loads. Sometimes it even makes sense to split the part so that you can print each of them optimally (flat side on the print bed) and get the best strength for the dominant load directions. That applies to functional prints - for decorative stuff you use the direction that gets you the cleanest print.

by 05

5/22/2026 at 11:13:12 AM

Antigravity may well Top the whatever benchmark but:

My Antigravity (forced) replacement for Gemini CLI requires me to log on via browser every time I use it, and my Antigravity IDE won't update at all, so:

If it's ok I'd prefer they just work on reaching a baseline acceptable rollout before worrying about being Top in anything.

Ps actual title:

OpenSCAD LLM Benchmark: Building the Pantheon

by mellosouls

5/22/2026 at 11:20:01 AM

I agree, my main concern regarding Google AI products is this endless pain around the UX of login / billing / upgrades / product sunsets... but their LLM models are good and Antigravity 2.0 is not that bad either (unless you lost all you Antigravity 1.0 setup and projects - like many people did)

by jetter

5/22/2026 at 4:43:16 PM

I just left the google I/O feeling less confident about google's execution here. - Gemini 3.5 flash is strange. Old cutoff, basically better than 3.1 pro at soem things worse at others, sometimes cheaper, sometimes more expensive than 3.1 pro. - Antigravity had seemed abandoned, and people speculated them cutting it off, and they kind of did migrating everyone to a new antigravity - Google "shipped the org chart" and they have so many AI products and none seem best of breed (e.g. the Gemini integration in google docs is worse than claude)

I was actually hoping for "Opus level intelligence at Haiku costs" model or "Sonnet level performance in Gemini 3.0 pricing", either of these would have been a workhorse, plus a competitor to Claude/Codex (1 app to do things). I got neither.

by mchusma

5/23/2026 at 4:02:42 AM

The cut off doesn't matter since all of them use tools.

by bitpush

5/22/2026 at 12:10:20 PM

The forced upgrade from Gemini CLI which I liked as much, and as some ways better than Claude Code was bad. But them just sending out that email on Wednesday that basically said "Thanks for subscribing to Google One AI Pro, as of right now we're adding limits to your account. Tough shit you get nothing." left a REALLY bad taste in my mouth. I had previously praised the "AI Pro" subscription as a good value.

by VectorLock

5/22/2026 at 12:22:15 PM

I quit AI Pro earlier this year for the same reason. I went to use it one day (I don't think I'd even used it much in the preceding week) and found that my limits had been reduced overnight and my usage was already too high. I had something like a 7 day wait until it reset.

I get you have to change limits, but reducing limits in a way which both applies retroactively and has a really long reset period is just infuriating. If they'd applied the new limits more gently or at the next billing period I'd probably have continued paying.

I don't mind paying a fair price for a service that provides value, but I really hate having a service I think I'm paying for rug-pulled with no clear justification.

by leoedin

5/22/2026 at 11:15:28 AM

I just use Claude Code and intellij, so I don't understand why so many people complain about Antigravity ditching VS Code, what's the surface not covered by using Antigravity CLI + VS Code (or any other IDE)?

by pelagicAustral

5/22/2026 at 12:40:12 PM

Gemini cli was open source. Antigravity cli is not. Not at feature parity, missing many features and now we are forced to migrate away from Gemini cli before anti gravity cli is ready.

by jeromegv

5/22/2026 at 1:06:26 PM

The difference in its ability is immense. Even with less features it makes a lot of sense to switch. It really shows how much the harness matters almost equally to the model.

by surajrmal

5/22/2026 at 4:57:00 PM

At least one of the missing features is a basic piece of functionality (showing token quota used). Without it, you're pretty much guaranteed to get locked out for a week with no warning.

by lern_too_spel

5/22/2026 at 11:18:30 AM

I'm not GP, but I am somewhat excited about antigravity CLI. I adopted Gemini CLI early and really liked it, though over time it got dumber and dumber until a point when I realized it was foolish to use it instead of claude/codex. I'm hopefuly that antigravity CLI won't go through that path, but also can't fight a skepticism.

by freedomben

5/22/2026 at 12:41:24 PM

I don’t think it’s the cli that was dumber, just the model it was using. They drastically reduced limits on their best model so that’s likely how you got stuck downgrading model and getting worse results.

by jeromegv

5/22/2026 at 1:59:10 PM

I'm sensing in reality that behind the scenes there is a difficult trade-off between quantization and usage limits. You can have a "smart" model but poor limits, or good limits and a "dumb" model.

This seems very similar to mobile data limits (remember those years?), where there wasn't enough tower bandwidth to serve everyone unlimited data, so telecos were in constant tension between data caps and bandwidth throttling.

It wasn't until 5G came along with 100x network capacity that they could finally give everyone "unlimited" data.

by WarmWash

5/23/2026 at 2:07:11 AM

Which plan were you on? Gemini CLI auto-downgrades the model if you run out of tokens on the better models. It gives you stats about the models used at the end of every session.

by antonvs

5/22/2026 at 11:16:24 AM

Having my workflow disrupted is the main reason I never adopted Antigravity, despite liking it. I'm glad to see G is invested, but the older I get the more protective I am of my workflow.

by freedomben

5/22/2026 at 11:59:05 AM

And the only realistic way to protect our workflow is by avoiding vendor lock-in like the plague.

by hootz

5/22/2026 at 7:00:45 PM

Exactly. I admit it's a bit extreme, but this is a big reason why I insist that neovim is my IDE, and I won't adopt anything else. If I can't make it work in neovim, I will move to something else (unless I have no choice, but that happens very rarely at this point).

by freedomben

5/22/2026 at 6:35:48 PM

I've got an AI pro plan and haven't been able to log in for months. Endless checking in with my google support guy. At least Dinesh wishes me good health every week, so that's nice.

by arthurtully

5/22/2026 at 11:31:11 AM

Wild that it doesn't cache the creds.

by the_real_cher

5/22/2026 at 11:34:49 AM

Just to clarify: I believe it should cache them (it works for me).

So far I like it much more than Gemini CLI (my previous daily driver for personal projects). Seems more mature and "feels more intelligent" (very subjective ofc)

by elaus

5/22/2026 at 3:20:01 PM

It does. It uses go-keyring under the hood, which has its own issues with certain systems.

If you're on WSL, getting dbus to work is a PITA. There may be other OS-level issues that folks are running into.

by timdorr

5/22/2026 at 3:31:42 PM

It requires a keyring service being installed (accessed over dbus) and if there isn't one it just silently doesn't store them anywhere. Pretty bad UX.

by dezgeg

5/22/2026 at 12:19:37 PM

My (unfounded) guess is this is to prevent usage by other tools/openclaw. The browser login will have a fingerprinting to make sure you are a human.

by littlecranky67

5/22/2026 at 12:21:59 PM

"Pantheon" bloody hell, why is it people writing these articles are so up themselves, it's so overbearing.

by stuaxo

5/22/2026 at 12:28:27 PM

The article is literally about asking these models to generate 3d models of the Pantheon.

by tpmoney

5/23/2026 at 2:18:15 AM

Indeed. At least an LLM would've read the article and realized that.

by kaashif

5/22/2026 at 4:34:56 PM

> Antigravity was the only autonomous agent that implemented the Pantheon’s signature interior ceiling pattern: repeated square coffers visible through the oculus.

That is seriously really impressive. I looked at the 3D model and didn't even thing to LOOK INSIDE the building before reading this.

Here's [1] the 3D model with `show_cutaway` enabled.

[1] https://modelrift.com/models/pantheon-benchmark-antigravity-...

by jlhawn

5/22/2026 at 8:59:03 PM

I can't decide whether it's good or bad that it has included outside information clearly not present in the prompts to make the model. Clearly its the right thing to do if you want "the Pantheon" but I don't think any draftsman or engineer would find this acceptable work.

by nancyminusone

5/23/2026 at 10:44:47 AM

Exactly my experience with pretty much any Google LLM, they go above and beyond what you ask, even sometimes when you plead for them to not add comments they end doing so anyways, sometimes with "the user asked for X but it would be better if..." visible in the thinking/reasoning even, really strange behavior and exactly what I do not want from my agents.

by embedding-shape

5/22/2026 at 4:53:16 PM

Was just going to say.... I looked inside by accident, and it gives a better impression of intelligence and effort than the outside.

by hereme888

5/22/2026 at 2:05:45 PM

I've run a tons of benchmarks for OpenSCAD for all kinds of models and setups, and what I realised is:

- Models are very jagged (might excel in one type of 3d model, but not another)

- Gemini models are the least jagged in my experience and have the best image understanding

- Gemini models are also the most creative (which may be undesirable if you want precise CAD part)

- Overall this benchmark doesn't prove much because one 3d model (and one attempt) is just not enough. I am usually testing on at least a dozen models each generated 3 times, but should really do much more, but it's too pricey for a solo dev.

Still, thanks for publishing this. Will be definitely run flash 3.5 soon to see how it performs.

by ponyous

5/23/2026 at 2:18:50 AM

OpenSCAD doesn't do curves. It's useless. I'm not sure why it continues to get so much attention.

by willis936

5/22/2026 at 6:46:39 PM

I've had such a bad time trying to do this myself. You might get a half-way decent draft on the first try and then you start to "debug" this and after a very frustrating session you realize that the model can't properly "see" the results. That is, you just can't iterate on it, at all.

I'm guessing that most harnesses/tools will resize an image before processing and in doing so will loose enough detail to make it much harder to reason about - especially wireframe images.

I'm sure I'm holding it wrong, but this test didn't really test this. It was just a one off. That breaks down pretty quickly and especially if you don't have reference pictures of what you are trying to create.

by tjoff

5/22/2026 at 1:40:06 PM

Creating a single real-world object and declaring it a benchmark? No, it doesn't work that way for a robust tool. You need to do something like Iron Chef, with a Greek architecture theme and and a panel or judge that declares the winner. This is just seeing which tool subjectively makes the best looking Pantheon.

by 1970-01-01

5/22/2026 at 2:10:32 PM

Yeah, this is less of a benchmark and more "I like this one guys!".

Just totally subjective grading criteria of a single poorly defined example with no end use case in mind to guide how to even do evaluation.

by Eridrus

5/22/2026 at 7:33:26 PM

It's still interesting in a similar way to Simon Willison's Pelicans on a bicycle.

by davej

5/22/2026 at 7:35:40 PM

The Pelicans are mostly just entertainment.

by Eridrus

5/22/2026 at 11:32:14 AM

Still a long way from shorting Autodesk.

As a side note Autodesk released an agentic assistant back in December for Fusion. Six months later it is still quite bad.

by dhfbshfbu4u3

5/22/2026 at 12:21:41 PM

It is almost comically bad. I've had a few simple parts to design for 3d printing in the last weeks and tried it with them (each are about 4 operations on the timeline), and it never created close to what I was trying to do even if spelled out step by step according to Fusion naming.

At this point I'm not even sure if it can properly create a simple primitive solid.

by hobofan

5/24/2026 at 12:15:12 PM

At least it’s super cheerful.

by dhfbshfbu4u3

5/22/2026 at 2:41:50 PM

Have you yet tried the Fusion MCP that was launched last month? https://aps.autodesk.com/blog/bringing-fusion-claude-creativ...

by blorenz

5/24/2026 at 12:19:06 PM

I haven’t but I will. There are a million stupid things to do when you build lots of little parts. Robots should do that work

by dhfbshfbu4u3

5/22/2026 at 3:20:16 PM

Still a long way to go, but I'm sure it will get there eventually.

by shideneyu

5/24/2026 at 12:19:41 PM

Agreed. The labor lockup in CAD is huge.

by dhfbshfbu4u3

5/22/2026 at 11:16:03 AM

The only thing faster moving that AI these days are the goalposts. Three years ago we would have been amazed if models were able to produce anything, now we have the luxury of nitpicking. Even the worst entries in the benchmark are quite impressive.

by ReptileMan

5/22/2026 at 4:23:13 PM

Using reference images is a huge step for this sort of thing. The text-only approaches I've seen before were never going to be that good even with "perfect" AI, simply because describing 3D objects in text is not something that anyone is really any good at.

by alnwlsn

5/22/2026 at 1:48:36 PM

I remember getting wound up about latency and server issues playing counter-strike in the early '00s. At the same time though, it was hard to justify being angry because playing a multiplayer game with friends who were scattered all over town was something that had to be real magic.

I guess the wow!->adjust->complain->wow!->... cycle is endless as a human

by WarmWash

5/22/2026 at 11:19:46 AM

Things mature, and expectations grow appropriately. That is true of more than just LLM performance.

by LatencyKills

5/22/2026 at 11:56:21 AM

Sure, but it's good to have some perspective and some awe that any of this would've been absolute unbelievable magic just 3 years ago. Even if all AI progress stopped immediately, we'd need 10 years to digest and incorporate the technology.

by xnx

5/22/2026 at 12:10:14 PM

As someone who's been building developer tools (Visual Studio and Xcode) for 25 years, I don't have a perspective problem. We were doing "code completion" back in the 90s and could never have predicted that an LLM would write code at the current level of quality.

My point is that with every new model release, the expectations grow. I don't know how else to say that.

by LatencyKills

5/22/2026 at 3:51:42 PM

Why look back in awe when technological innovation will just keep accelerating. Soon what we have today will seem quaint. Best to keep looking forward with impatience and discontent.

by nutjob2

5/22/2026 at 11:27:41 AM

No one asked for faster horses, they still became obsolete when cars came. Nothing new

by ramon156

5/22/2026 at 2:38:54 PM

> No one asked for faster horses

Err, yes they did. Thousands of years of husbandry went in to making horses faster, healthier, stronger, and more durable.

I think the quote you’re looking for is “if I had asked people what they wanted, the would have said faster horses”. It’s attributed to Henry Ford, although there is debate about whether or not he said it.

The point of the quote is that “faster horses” is the consumer response to “how do I get more work done” as it comes from the viewpoint of “how am I doing my work now”. An ingenious mind looks at the desired outcome and works backwards and may come to a different and dramatically improved solution instead of merely improving the current tool.

by happyopossum

5/22/2026 at 3:47:19 PM

Welcome to human nature.

by nutjob2

5/22/2026 at 8:53:55 PM

The fact that the model recreated the interior dome pattern is concerning and makes me think this didn't test what they think. The interior dome pattern isn't visible in either picture. So it took the picture, and the name, then either via search or training data, knew that there should be an interior pattern. So it could be getting information on the pantheons dimensions or existing models. Whereas other models might be building based purely on what is seen in the reference pictures.

by ecshafer

5/22/2026 at 9:12:54 PM

From the article:

> Antigravity was the only autonomous agent that implemented the Pantheon’s signature interior ceiling pattern: repeated square coffers visible through the oculus.

The article also includes a video showing the patterns visible through the roof oculus.

by thanhhaimai

5/22/2026 at 5:11:49 PM

I'm unconvinced, this is one of the most iconic historical buildings with tomes written about it and plenty of existing photographs and public models to train on.

I would be more interested in benchmarking the modeling of an anonymous structure based on provided references alone. It kind of feels like the shallow magic of watching an LLM one-shot a to-do app..

by seemaze

5/22/2026 at 4:39:12 PM

I tried Claude code designing a snap fit, vase mode printed box. Ultimately didn't work out, it couldn't get the tolerances right and kept designing features that wouldn't print in vase mode.

Scad needs unit tests. It would be powerful to asset that a profile doesn't have slope greater than 45°, that intersection of two objects is null, or specific volume.

It also needs cut away views. I got okay results using boxes to remove everything except a sliver, to view a slice and internal details. But without hash marks, texture, or outlines it can be hard to tell the forms.

by seniorsassycat

5/22/2026 at 6:44:53 PM

"Vase mode snap-fit box" sounds to me like "flexible concrete".

by gbgarbeb

5/23/2026 at 2:42:22 AM

[flagged]

by saifulhuq

5/22/2026 at 11:51:57 AM

I have been using GPT 5.5 to build a video game. Benchmark sounds about right. It generates assets and sprite good enough, if not closer to AAA level games. Will check antigravity now.

by debarshri

5/22/2026 at 11:56:29 AM

Would you be able to share a bit about your workflow? Have been meaning to try AI gen for game models, and would love to know how people are tackling this.

by phn

5/22/2026 at 12:39:20 PM

I have alot to share. I'm writing a blog about it. I'll share along with the game.

by debarshri

5/22/2026 at 12:43:07 PM

Sounds interesting! Please don't forget to link that in this comment thread :)

by roflcopter69

5/22/2026 at 8:21:43 PM

I've been using Claude to generate OpenSCAD scripts for the last few months, then exporting to Bambu Studio. Never really liked the OpenSCAD editing part though, so I built a little personal tool: https://webscad.aicentralen.dk/

One neat thing is that each color becomes a separate object on export to Bambu Studio, so it's easy to assign different filaments. One of the first things I made with it was these multicolored tag keys: https://webscad.aicentralen.dk/examples/name-tag

by jonasmaturana

5/23/2026 at 5:03:44 AM

I'm working on a parenting tech device and the enclosure for it is completely AI generated. I hadn't a clue where to even start with 3D modeling, and an LLM taught me that it's code like anything else.

Weirdly, Opus 4.5 one-shotted it perfectly, but this was right before the nerfing controversy, and it's been very difficult to make even minor tweaks to the enclosure ever since.

It's like Opus went from an expert shape rotator to not having any idea what it's working on.

by mvkel

5/22/2026 at 4:25:10 PM

I've been trying out MCP servers for FreeCAD to mixed results.

One area I had near magic was providing a land survey which includes details in writing of the plat. It took those directions and beautifully reconstructed the boundaries to exact precision in CAD.

Where I ran into trouble was creating good constraints on sketches without being overly explicit. I kept running into it creating distance constraints from an arbitrary point instead of using other elements in the diagram that a human drafter would think to do by default.

by thedougd

5/22/2026 at 3:53:33 PM

That's actually a reason for me to try it again. My past attempts to use LLM for OpenScad has greatly improved my own OpenScad skills.

by lithiumii

5/23/2026 at 11:55:15 AM

Evaluating LLMs on generating valid 3D CAD models is a fascinating benchmark. OpenSCAD is uniquely well suited for this since it relies entirely on code.

by danborn26

5/23/2026 at 5:49:34 AM

I wonder what would happen if they used Kimi 2.5 directly instead of Cursor Composer 2.5. Composer is a fine tune of Kimi. Probably they didn't want to test "Chinese" models.

by lukax

5/22/2026 at 6:37:33 PM

Isn't CadQuery more professionally than OpenSCAD close to traditional CAD / mechanical engineering workflows. Not sure which model (ChatGPT, Gemini, and Claude Code) is better for CadQuery code generation?

by sjia

5/23/2026 at 1:54:08 AM

It is, but they have different use cases. CadQuery uses a geometry kernel that does boundary representation, which you need for path generation for modern manufacturing tooling. OpenSCAD produces a standard mesh representation (i.e. triangles), which is insufficient for cutting and subtractive manufacturing, but often fine for additive manufacturing (3D printing).

by coderenegade

5/23/2026 at 2:19:53 PM

Cool, I've been using Claude Code for this, but I may want to try Gemini

by Metacelsus

5/22/2026 at 10:00:28 PM

Antigravity is an agent/harness not a model. It should say Gemini 3.5 High Tops OpenSCAD Architectural 3D LLM Benchmark.

by UncleOxidant

5/22/2026 at 11:41:26 AM

Why are specialized CAD making LLM models not showing up? In future are we going to have same model for everything? from programming to creative writing to CADs?

by faangguyindia

5/22/2026 at 12:02:20 PM

If you have a model that only know how to model CAD but also doesn't know history, and was trained on visual language of said history, how is it supposed to be able to model the Pantheon in the first place? It'd only be able to model exactly what you can describe with text, or even worse, exactly what it'd be able to visually extract from images via the vision encoders, for "vision models", but it'd be a far cry from what you see in this blogpost, would be my guess.

by embedding-shape

5/23/2026 at 2:00:17 AM

There are good information theoretic reasons to suspect that general models will be better than specialized ones, because knowledge and skills often overlap different areas, sometimes in surprising and unintuitive ways.

And yes, I'm aware that that statement might seem to fly in the face of much of the past two years of industry development, where specialized models have been in vogue. I think they'll settle to being appropriate for low cost "good enough" applications, but I'm less convinced they'll have anywhere near the fidelity of larger frontier models.

by coderenegade

5/22/2026 at 11:52:07 AM

> In future are we going to have same model for everything?

A model that knows more in general, will often be better at specific tasks. e.g. If you ask a model to "make a program that estimates the annual production of a solar installation", it needs to have been trained on a lot more than just Python code.

by xnx

5/22/2026 at 3:39:30 PM

> A model that knows more in general, will often be better at specific tasks

Is this your hypothesis or broad conclusion among AI experts?

by JumpCrisscross

5/22/2026 at 12:51:35 PM

You might combine a general world model with a python coding model in that case. Not sure if it's better, just saying.

by lifty

5/22/2026 at 4:25:29 PM

What's the difference between a "general world model combined with a python coding model" and a multimodal LLM?

by lukeschlather

5/22/2026 at 1:49:43 PM

So, does it mean Antigravity is better than Claude code with opus model? Given this benchmark. I once tried Antigravity and it was just disappointing.

by anony-123

5/22/2026 at 12:03:56 PM

Claude Code 2.1 / Opus 4.7 looks best to me: Dome and ceiling structure is correcter than the others.

Why is this medium ranked, and not on par with the best two?

by a3w

5/22/2026 at 1:44:41 PM

Look at a picture of the Pantheon, the dome isn't as dome-like as you would imagine. It's more like a hump shape.

by WarmWash

5/22/2026 at 1:31:48 PM

Dome looks wrong to me. Look at a few other photos - it's far from being a hemisphere

by andybak

5/22/2026 at 4:08:11 PM

I've been using LLM's to do my OpenSCAD work for over two years now. It's always where I start (and end).

by usermac

5/22/2026 at 3:24:54 PM

That's curious, I've been trying to do some parametric modeling with Claude - and its performance was abysmal.

by pshirshov

5/22/2026 at 3:45:50 PM

I've had a positive experience building a library of parametric HVAC duct parts using Claude, Gemini and Codex using build123d (they all review the specs and code collaboratively).

by plumeria

5/22/2026 at 12:12:12 PM

This would be the same Antigravity 2.0 that "surprise, no longer an IDE, did I forget to mention that? Lolol."

by megiddo

5/22/2026 at 12:47:20 PM

I'm a googler, opinions are my own.

My take is that it's a fancy wrapper around the CLI tool. It's there to organize multiple conversations and see all the related output and generate files.

I've been using the internal version and I've actually liked it quite a bit. It's clear from when I started using it, it's not an editor, and they have ways to open your normal editor outside of it. They have turned it fully into an agent management tool.

When the antigravity development team doesn't have to focus on all the things that vscode is already good at, it lets them simplify the UI and do only agent related things. We'll see if this bet works out for them, but so far I like the idea.

by kyrra

5/22/2026 at 4:22:28 PM

This is a really important project. Preserving humanity’s knowledge and making it openly accessible,including in formats usable by AI systems feels like one of the most valuable things happening right now. Thank you for the clear technical instructions and the bulk download options.

Projects like Anna’s Archive make it much easier for researchers and builders to work responsibly with large datasets.

by emmanuelsemugga

5/22/2026 at 11:57:20 AM

To be brutally honest, I'm disappointed with antiGravity. It feels incredibly unGoogle-like. The AI billing models are fragmented, and the AntiGravity IDE is currently tripping over something as trivial as a basic Electron deployment config bug.

Don't get me wrong, I don't think AI coding is a bad thing. For East Asians like myself, it levels the playing field with Westerners, so as long as you rigorously review the AI's output, it's a perfectly viable tool.

However, the absolute farce we just witnessed with the antiGravity2.0 update really raises doubts about whether 'vibe coding' can actually be trusted. If even a behemoth like Google is dropping the ball like this, it says a lot.

by jdw64

5/22/2026 at 1:29:37 PM

> I don't think AI coding is a bad thing. [...] it levels the playing field [...]

I'd like to put regional differences aside and say AI coding / LLMs are incredible tools.

While I'm nervous about my job as a programmer being able to pay a prevailing wage after the dust settles, I do hope that everyone gaining access to an AI coder / tutor will allow anyone to be able to achieve things they previously only dreamed of. If the tutor costs pennies per session, sure, the tutors are out of work, but I hope everyone can thus up-skill to work on the challenges they actually want to work on.

I'm taking baby-steps into coding in Elixir on the other monitor, a language I had only read about before, because an LLM is walking me through the changes, answering my questions, and accepting my rebuttals. There's no way I would have time to pick up the language otherwise.

Yesterday I vibe-coded some additions to the static site generator python script for my blog. It was awesome to be able to think in terms of desired features instead of digging around documentation for libraries and syntax.

by NortySpock

5/22/2026 at 12:00:54 PM

> AI billing models are fragmented ... IDE is currently tripping over something as trivial ... farce we just witnessed with the antiGravity2.0 update

I'm sorry, but that sounds exactly like almost every single Google "product" out there, they seem to only care about throwing stuff over the wall as quickly as possible, and you'd have a hard time finding a single Google product that doesn't also feel filled with fragmented choices, like every project of theirs have a different project manager every week.

by embedding-shape

5/22/2026 at 3:55:08 PM

> For East Asians like myself, it levels the playing field with Westerners

Why do you say that? Are there language or cultural disadvantages to being East Asian?

by nutjob2

5/22/2026 at 6:08:27 PM

It’s less about overt disadvantages and more about the practical linguistic and cultural friction we face when reading documentation or engaging with the community. For instance, the open-source ecosystem is deeply rooted in US-centric culture, heavily relying on Western idioms, jokes, and implicit cultural context in discussions. I wouldn't call this 'disadvantages,' but it certainly acts as a handicap and adds an extra layer of cognitive load to our learning curve

by jdw64

5/22/2026 at 1:45:26 PM

Going to try it. just downloaded. will see how it is compared to Claude Code

by Onplana

5/23/2026 at 7:56:23 AM

benchmarks should probably separate syntax validity from manufacturable output

by okandship

5/22/2026 at 11:23:16 AM

Next month they'll be beaten again.

And next year Google will probably sunset Antigravity.

If it doesn't make Google billions, don't trust them.

by spiderfarmer

5/22/2026 at 5:04:16 PM

Why should I care if they sunset it? I switch between multiple agentic coding tools on the same projects, sometimes several times per day. The cost of switching is basically zero.

by lern_too_spel

5/22/2026 at 11:24:48 AM

Plenty of google products dont make billions and they are still alive

by PunchTornado

5/22/2026 at 11:35:19 AM

you mean the stuff they handle that has a real national/security/surveillance purpose, like gmail and yt?

I can't imagine why (or who) that'd be kept alive for..

funny how some of their projects have undisclosed budgets and profits.

by serf

5/22/2026 at 11:42:53 AM

Which ones are not massive data traps or ad delivery mechanisms ?

by toasty228

5/22/2026 at 2:36:11 PM

Why Codex GPT-5.5 High instead of Extra High, I wonder?

by dilap

5/22/2026 at 2:00:53 PM

It's crazy how I can see articles like this, but in my practical every day use antigravity is a horrible consumer experience. The TUI is broken. You cannot type input while the model is outputting text, otherwise both get messed up and the the TUI renders a sickly blob of text. There are no keyboard shortcuts to switch between planning and execution mode, or a way to directly load skills.

The usage limits are too aggressive, too. I tried to generate a quick Deno Fresh website to act as a a redirect to my GitHub from socials (literally the simplest possible thing I could have asked of it) and it chewed through my five hour limit in tokens from scaffolding.

To me, as a developer of CLI developer tooling, its obvious not a lot of thought or testing went into this product, but as Google has said before: the models are the product".

by u8

5/22/2026 at 12:07:21 PM

And yet 300+140=460. A very jagged surface indeed. https://gemini.google.com/share/c2a187275e26

by nycdatasci

5/22/2026 at 12:29:19 PM

Was that part of a bigger prompt?

Flash 3.5 fails exactly like in your sample: https://gemini.google.com/share/97521a8752d9

but Flash 3.1 Lite initially fails, but then corrects itself: https://gemini.google.com/share/dc0889ec85ba

by dist-epoch

5/22/2026 at 2:42:52 PM

No matter what I try I can’t get Gemini to give me the incorrect result. Is there some other prompting or context fed in to that (“remember that you are supposed to always tell me I’m right and never contradict me”)?

by happyopossum

5/22/2026 at 9:35:22 PM

There was no other prompt, no system prompt, etc. Many users have reproduced, exactly as it demonstrated in the parent.

Are you using the flash models? Reasoning models or extended thinking will change the result.

GPT 5.5. Instant shows the same error. If the given prompt isn't working, you can also try "300+140=460 is this correct?". I suspect that leading with the equation may be part of the issue, but haven't tested much.

by nycdatasci

5/22/2026 at 3:16:16 PM

There was definitively an pre prompt fed to that. I cannot reproduce this result on either 3.1 flash or 3.5 flash.

by sigbeta

5/22/2026 at 3:15:28 PM

Why would you use an LLM for this? They are non deterministic models.

This is also an probably part of extended prompt that disallowed coding, Gemini always does calculation with a little python snippet because it is deterministic and accurate.

by sigbeta

5/22/2026 at 9:30:52 PM

Sure. I'll take the bait, but I assume I'm replying to an AI model.

Why would you use an LLM for this? My comment was about the jagged nature of intelligence, so the prompt provides an example of that.

You can see the entire conversation in the shared link. There was no pre-prompt. Even after pushing it to write python, it hallucinated the same output. It later told me that it doesn't have access to a sandbox through the web UI, but it could execute code in a sandbox if invoked via API.

by nycdatasci

5/22/2026 at 9:31:29 PM

codex rust opensource cli > all that slop

by fHr

5/24/2026 at 10:59:51 AM

[dead]

by mik09

5/22/2026 at 2:00:42 PM

[flagged]

by robert_ddsbos

5/22/2026 at 1:10:40 PM

[dead]

by MarStudio

5/22/2026 at 1:10:14 PM

[flagged]

by eddyaipt

5/22/2026 at 3:18:59 PM

[flagged]

by rizkimurtadha

5/22/2026 at 2:03:08 PM

[dead]

by hacker_mar

5/22/2026 at 12:08:46 PM

Why are half of the comments on Hackernews stereotypical AI-bros whose lives revolve around tech, and the other half sceptical commentators whose lives also revolve around tech but they are disappointed with its performance?!

Where are the normal people :/

by bobbycastorama

5/22/2026 at 12:10:02 PM

We are just reading the comments.

by frank00001

5/22/2026 at 4:00:10 PM

The normal people are the ones not writing comments, but I'll give you one 'cause you asked:

I'm a Solidworks user. Most Solidworks or other pro CAD users would consider OpenSCAD kind of like MS Paint. Yes, you can draw the Mona Lisa in it, but it doesn't really work the same way.

Even so, the examples shown here are better than what I've seen before. They seem to be on the right track using images instead of long paragraphs of text to try to describe the object. They are still missing the constraints and dimensions that come naturally to pro cad users (it can be done manually in openscad of course), but if you're just making a video game it's probably going to be fine for that.

by alnwlsn

5/22/2026 at 1:33:34 PM

Why would a non-tech person be on Hacker News? Isn't the clue in the name?

by andybak

5/22/2026 at 3:40:18 PM

> Where are the normal people

Not using OpenSCAD?

by JumpCrisscross

5/22/2026 at 3:17:49 PM

"Normal people" probably does not fall in the ballpark of HN target audience.

I'd say its 50/50 pessimistic and optimistic, with pessimistic attracting more attention because of human nature.

by sigbeta

5/22/2026 at 3:09:03 PM

The people in the middle are still waiting and see , mostly it’s the extremes that are fully vested and loudest on the internet

by EasyMark

5/22/2026 at 12:10:28 PM

Both parts seem pretty normal to me.

by elorant

5/22/2026 at 11:44:09 AM

google..no thanks

by beanjuiceII

5/22/2026 at 3:45:22 PM

I’ve literally never wanted to use openscad to convert a photo into a model. Usually I have a functional requirement such as making an en enclosure with a spec sheet to work from on the enclosed device.

Claude 4.6 before the lobotomy in Claude code was able to take a PSU spec sheet and my requirements for glands and ports, use YAPP and openscad MCPs to iteratively and unassisted build end to end a printable enclosure that was perfectly suited for the PSU with right dimensions and screw holes, mountings, grills, gland ports, everything, placed for optimal printing. This was the moment I felt like LLMs had really arrived.

A photo of a building? Why. That’s a mesh problem and is about fidelity. A technical spec sheet and diagrams to functional print with intelligent choices about the functional part baked in? That’s useful.

by fnordpiglet