Runway Wants to Own the Fourth Dimension

Runway just dropped General World Models and the framing is, characteristically, maximalist — not "we made the video better," but "we are going to build a compressed representation of reality."

The thing is, they might have to.

Text-to-video is getting solved. Not perfectly, not by everyone, but well enough that it's starting to feel like a features race — more seconds, higher resolution, better motion coherence. That's not a moat. That's a spreadsheet. Runway raised hundreds of millions of dollars and they did not do that to win a spreadsheet.

So the bet becomes: what's actually hard. And what's actually hard is that none of these models — not the video ones, not the LLMs, not whatever Google was pretending to show us last week — provably have an accurate world model. They generate plausible surfaces. They do not know that a glass falls when you let go of it because gravity is a property of mass, not a pattern in training data. That distinction matters more the longer your time horizon gets.

The Gemini demo situation is instructive here, actually. Everybody saw something that looked like real understanding, and then everybody found out it was staged, and now nobody agrees on what any of these systems actually understand versus what they've memorized the shape of. Which is a useful state of uncertainty to be in when you're about to announce that you're going to build the real thing.

If Runway cracks world models, video generation stops being their product and starts being their demo. Four dimensions — three spatial, one temporal — understood well enough to simulate forward. That's not a video company anymore. That's something else, and nobody's quite got a word for it yet.

They've got the money, the model, and the timing. Whether they've got the idea right is the only question left.