Skyfall AI

An AI lab advancing intelligence through world models

[Skyfall AI @skyfallai]

Today we're unveiling SKY 00-1, the first Engineering World Model towards Autonomous Manufacturing.

LLMs are great at information retrieval and code, however, the physical world involves geometric and spatial understanding. Designing complex physical objects requires long-horizon planning and latent reasoning, which LLMs weren't built for.

We stacked SKY-001 against the top frontier LLMs (GPT-6 Astra, Fable 5.1 etc) and we found that our model is 100x faster and 150x cheaper in CAD and BIM while having higher precision.

As a compositional visual world model, SKY-001 lets you "dream" a design before you build it. It explores many designs in parallel, breaks each one down into its smallest components, and reassembles them in 3D, ready to open in CAD tools like Autodesk Fusion.

With applications in several industries, SKY-001 is the first step to reimagining engineering firms.

👉 Join the frontier with us and visit our website for early access (link in comments).

[Sam Pasupalak @spisallyouneed]

Introducing SKY-001, the first Engineering World Model.
On CAD and BIM tasks, it beats GPT-6 Astra and every frontier model we tested:

-> 51-125x faster
- > 150-210x cheaper
-> Higher IOU than all frontier models

SKY-001 lets you dream a physical object before you build it.

18 months of work. Built by the team out of Maluuba's deep learning lab (acq. by @MSRResearch).

The first ‘AI General Engineer’ will be the backbone for Autonomous Manufacturing.

[Skyfall AI @skyfallai]

@aryaan_smm @spisallyouneed More coming soon, stay tuned!

[Sumit Pasupalak @omgiamgod]

Grateful to be tackling some of the hardest problems in AI with such an incredible team. Couldn't ask for a better group of people to build with.

[Skyfall AI @skyfallai]

Full technical breakdown in our blog: https://t.co/wR8afQOlXa

[Skyfall AI @skyfallai]

Meet the Research Team behind SKY-001 💙

Sign up for the waitlist if you haven't yet: https://t.co/tjtpo9VJE5

[Sam Pasupalak @spisallyouneed]

Meet the world-class research team behind SKY-001, the first Engineering World Model.

Solving complex design and engineering problems in the physical world is orders of magnitudes more difficult than LLM agents using next token prediction. Our team at @skyfallai decided to take a different path than everyone else out there in order to achieve the research breakthrough.

Last week, we shared how SKY-001 stacks up against frontier LLMs, showing the clear advantage World Models have on complex real-world tasks like engineering design.

Here is a sneak peek video where you can learn directly from the researchers about what makes our approach unique, what training and data collection paradigms were used, and why we believe World Models are the next frontier in AI.

I’m incredibly proud to work alongside this team.

[Rohan Paul @rohanpaul_ai]

World models are moving out of robotics and video and into engineering software.

SKY-001 from @skyfallai turns a text prompt or a 2D sketch into a full CAD build sequence that opens directly in Autodesk Fusion.

hear directly from the research team behind SKY-001"

Speed separates them, with each case taking 3.9 seconds versus 199 seconds for GPT-6 Astra.

SKY-001 gets there by planning visually instead of through repeated LLM calls, first splitting an object into sub-assemblies and individual parts.

A construction world model then predicts the ordered sketches and 3D solid states that build each part, which LLM-made shapes usually lack.

[Skyfall AI @skyfallai]

Full technical breakdown in our blog: https://t.co/wR8afQOTMI

[Skyfall AI @skyfallai]

Here's a side-by-side comparison for BIM 👇

Which domain should we take it to next?

Get early access: https://t.co/tjtpo9VJE5

[Sam Pasupalak @spisallyouneed]

Our vision at @skyfallai is to make intelligence as affordable and ubiquitous as electricity.

With SKY-001, we're one step closer to that mission. Here's a side-by-side comparison of how fast our World Models are compared to frontier LLMs in CAD. LLMs are too slow so we had to 10x the video speed gosh 😂.

Electricity became ubiquitous because it was both cheap and fast enough to be woven into every tool and process without a second thought. When engineering design generation becomes 1000x faster, it transforms into a real-time utility for the entire human race.

[Skyfall AI @skyfallai]

@okxuzb @spisallyouneed did you try here: https://t.co/tjtpo9WhtD?

[Skyfall AI @skyfallai]

@blackblexbladbl @spisallyouneed join us at the frontier: https://t.co/tjtpo9VJE5

[Skyfall AI @skyfallai]

@stevencheng 👋

[Skyfall AI @skyfallai]

@Alwurts yes! that's exactly what it is designed to do.

[Skyfall AI @skyfallai]

We launched the first Engineering World Model, SKY-001, this week and it already outperforms all the frontier LLMs in CAD and BIM in speed, cost and precision.

Here’s an example of how our model compares to OpenAI’s Astra, Anthropic’s Fable 5.1 and Google’s Gemini 4.

These results did not happen by accident. Read our technical blog to find out how we built SKY-001 (🔗 below).

[Skyfall AI @skyfallai]

@OnPremConnor We're working on it and will share updates soon. Join our waitlist: https://t.co/tjtpo9WhtD

[Skyfall AI @skyfallai]

Sign up for early access: https://t.co/tjtpo9WhtD

Blog: https://t.co/BITBrf1dA0

[Skyfall AI @skyfallai]

👉 Generate your own game now for @arcprize.

[Sam Pasupalak @spisallyouneed]

The @skyfallai team has created a AI generator that makes ARC-AGI-3 styled games end to end with no human involvement

All games in this repo have a formal spec, solver trajectories, per level budgets and multiple layout seeds. We also have a bundled browser player to go through these games.

The generator comes up with ideas (follows a design vocabulary), creates detailed specs that are then implemented and verified (by a second agent) in multiple iterations and finally packaged in the format above.

You can run this generator locally to create any number of original games to help train/evaluate your agent.

We have taken some inspiration from @tufalabs re-arc-3 to create this generator. Please feel free to try this out, share your feedback, and contribute on this to help the @arcprize community.

Repo: https://t.co/pLDxi9M7Q5

[Skyfall AI @skyfallai]

@logicwins awesome!

[Skyfall AI @skyfallai]

👉 Sign up for early access now: https://t.co/tjtpo9WhtD

[Sam Pasupalak @spisallyouneed]

While designing complex physical objects via CAD, frontier LLMs try to brute-force engineering through text tokens and inevitably buckle under compounding error pile-up.

SKY-001 operates as a true spatial World Model. It plans directly in the 3D space, recursively breaking complex designs down to the smallest part, then re-assembling everything while keeping every step globally consistent.

That’s how it leaves GPT-6 Astra, Gemini 4 Argon, and Claude Fable 5.1 behind on complex CAD workflows.

Here’s a side-by-side look at a Jaguar E-type sports car and a Ducati Cafe Racer:

[Skyfall AI @skyfallai]

We took a contrarian approach in research, and we're glad we did. The results beat our own expectations.

SKY-001 outperforms all frontier LLMs such as GPT-6 Astra, Gemini 3.8 Flash, and Opus 5.5 in both CAD and BIM. It's a big step forward for world models proving how such AI systems can reinvent manufacturing firms.

Next, we're excited to bring it to more domains such as PCB Design, Aerospace Engineering, Computer Use Agents and many more.

Join the waitlist now for early access: https://t.co/iQZCW8zSSQ

[Skyfall AI @skyfallai]

Blog: https://t.co/BITBrf1dA0
Website: https://t.co/tjtpo9WhtD

[Skyfall AI @skyfallai]

Here is how SKY-001 beats prompt-based methods step-by-step:

Stage 1: Parallel Explorations Through World Modeling
Stage 2: Recursive Diffusion-Based 3D Decomposition
Stage 3: Multi-step 2D to 3D Latent Action World Model
Stage 4: 3D Construction World Model

Check out our full blog post for all the technical details! 🧵