China Is Ahead on World Models for Robots | Anastasis Germanidis artwork

China Is Ahead on World Models for Robots | Anastasis Germanidis

MTS

November 12, 2026

Runway co-founder and CEO Anastasis Germanidis explains why the video generation lab open-sourced Praxis-1, its new world action model topping NVIDIA’s RoboLab 120 benchmark.

0:00

Anastasis Germanidis

I think China is further ahead than the US in terms of being world model build, of just like really understanding how powerful world models can be, and making sure that the West is also competitive on that front is really critical, and making sure that we open up the research so that others can contribute to it and make those models better.

0:23

Sophia

Hello everyone, welcome back to MTS. Today, we are joined by Anastasis, co-founder and CEO of Runway. This morning, Runway released Praxis-1. It's world action model for robots, with open weights for anyone to use. It turns Runway's video pre-training into control for real robots, and it's at the top of Nvidia's RoboLab 120 leaderboard. So Anastasis, welcome to MTS.

0:49

Anastasis Germanidis

Good to be here. Thanks for having me.

0:51

Sophia

Great to have you. Big news for you guys today. In your words, tell us about the announcement. Why is this a notable moment?

1:00

Anastasis Germanidis

Yeah. So Praxis-1 is essentially the culmination of a year-long research effort to start from all the work that we've been doing over the past almost half decade at this point on video generation models and scaling video generation models and applying that work into robotics. So Praxis-1 is essentially a world action model. So it predicts motor commands and controls robots for a variety of manipulation tasks. We've seen it performs quite well.

It's at the top of the leaderboards, as you mentioned. It's able to do a bunch of wide range of manipulation tasks. It's able to adapt to new embodiments very quickly. So we're very excited about it.

1:43

Sophia

One of the things I wanted to ask you about is why video? People know video is the bread and butter of runway and all the models that you guys are building. Tell us a little bit more about why video for Praxis-1.

1:56

Anastasis Germanidis

So the main idea is, what are you essentially doing when you're training a model to predict pixels and predict video?

You're learning about how the world works. You're learning about physics, you learn about how do humans perform different tasks out in the world, how they play sports, perform household tasks. So the main intuition is, if a model becomes really good at predicting video, predicting all those observations from the world, then there is no reason it should also not be able to predict the same thing but in the action format that's compatible with something that you would use to control a robot. So the main thesis of Runway is, if you train models that are good enough at simulating the world, then they're going to be useful obviously for media and content creation and marketing and advertising, which has been the bread and butter of Runway. But they're also going to be useful for any kind of system that interacts with the physical world, because during pre-training, they've learned so much useful stuff about how the world works.

2:57

Sophia

Yeah. So tell us a little bit about how it's already being used earlier today on MTS. We showed some videos of robots that were using Praxis-1.

Tell us a little bit more.

3:09

Anastasis Germanidis

Yeah. So at this point, we've adapted Praxis-1 to nine different robotic hardware. So we've worked with a wide variety of partners from humanoids for construction, with our work with noble machines to buy manual robots, to robots for manufacturing, to we're doing some work with Ultra on logistics. So the range of different tasks that the same model is able to do is quite wide.

And we're also now using it in the office in a bunch of different ways. So it's been interesting to see over the past year kind of a huge software company kind of convert itself to a hybrid software and hardware company.

3:56

Sophia

You've been using it at the office? How so?

4:00

Anastasis Germanidis

Yeah, so essentially just like on simple tasks like sorting swag or kind of helping with very tasks around the office. It's been a really nice way to just test kind of how our own model is useful for the things that we care about.

4:17

Sophia

What's it been like for everyone who works at Runway? I know typically it's like, okay, we can now generate video for filmmakers. But now for all the employees, they see physical robot arms dealing with swag. What's that transition been like transitioning a lot more into hardware in the physical world?

4:35

Anastasis Germanidis

I mean, I think the only constant of Runway has been changed. We started eight years ago when nobody really cared about generative models and we had to transform ourselves into being a frontier research org that's able to pre-train video models at scale, which was a big transformation for the company. And now that's just another transformation that's been happening over the past year. And so I think the same lessons of how we built up the company, the way kind of thinking of things in terms of like first principles, that kind of creative mindset really applies to the work that we're doing in robotics as well.

Read the other 8 minutes of this episode here.

$15 gets you this episode and 99 more. Credits never expire. After checkout, one link brings you back here to read it.

Not what you expected? Full refund within 14 days.

Thousands of transcripts fetched by people building searchable podcast archives

Fetch transcripts with code

The same credits work from a script or an AI agent. The demo key returns a sample episode in full, no card needed:

request
curl -H "x-api-key: pt_demo" \
  https://spoken.md/transcripts/1000651996090

Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.

From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.

Using your own key:

request
curl -H "x-api-key: YOUR_KEY" \
  https://spoken.md/transcripts/1000801523995