It’s time to go beyond language models. Introducing Odyssey-2 Max, our most powerful world model yet. It materially advances the SOTA in physical accuracy. This is a big step toward models that simulate and interact with the world in real time. A new intelligence entirely!

Enjoy our technical report on Odyssey-2 Max here, and apply for access!

Physics!

This is one of the most mind-blowing outputs to me.

$800K TC Jane Street Quants realizing they could’ve used their physics background to cook up world models instead of trading soybeans for 50 years:

Seeing these reflections generated in real-time is mindblowing

This one too!

So excited to see fluid dynamics being properly modeled in an autoregressive fashion ☺️

🥹

we just went from predicting words to predicting reality. here is why that changes everything.

That's a great way to put it.

means a lot coming from the team building it. the applications you are going to unlock with this are going to be wild to watch.

Do this now

The future

This is insane. We have come so far, I remember seeing early itterations of this a few years ago. We have come so far.

Still really in the equivalent of this for early gaming.

So awesome to see scale can make all the difference between physics that can pass for looking reasonable to one that is super realistic!

Scale is all you need (sorta).

you guys cooked here ngl

We can also simulate cooking.

Finally someone focusing on physics instead of words. Well done.

The shift from describing the world to simulating it is starting to show.

Ok, but can we at least point the bus the right direction?

Is this trained on any copyrighted work without a licence?

let’s gooooo!

Are we in the simulation now?

Is the simulation in the room with us.

🤔🤔🤔

"We are living in a computer-programmed reality, and the only clue we have to it is when some variable is changed, and some alteration in reality occurs." - Philip K Dick

⏰⏰

Omg I love this!! Team has been cooking!! 🧑‍🍳

Inspired by you 🤝

nah, you're goats

imma cook something with this soon 🤭

Ready Player one vibes

really validating to see how we get better visual quality + soft body dynamics from scale, all while keeping it real time!

So just hypothetically, what would it take to create a whole simulation world now? A Dyson sphere for power, a localized LLM in each generated rendering of an object, 3-D plane awareness, localized cache to only build what the simulations are looking at or measuring, connect them all, and then….. a whole world

I’m so so so impressed

I'm looking forward to checking this out. I do think we're on the way towards world models and away from language models for the next phase of artificial intelligence. Where there's deeper comprehension beyond simply the inferred relationship between words and other data.

Can't wait to show everyone!!

sooo luv it I can't wait for your model to be used for modeling and filmmaking.

Damnnnn

exciting times ahead for world building

Feels like we’re moving from talking about the world to actually simulating it.

Congrats on the launch

Impressive tech. Biggest issue for videos has been editing an existing clip. Glad to see how far it's come

Now imagine this with Odyssey-2 Max!! 🔥