Runway Gen‑4.5 ships 10s 720p clips – GWM‑1 previews live worlds

Runway turned its Research Demo Day into a real product push: Gen‑4.5 is now the flagship text‑to‑video model for all users, outputting 16:9 720p shots at 5, 8, and 10 seconds and topping Runway’s internal Elo board at 1,247. Native audio and true multi‑shot timelines are labeled “soon,” which matters because it points to a single Runway stack where images, camera moves, and soundtracks live together instead of being stitched across three different tools.

The bigger swing is GWM‑1, an autoregressive “General World Model” built on Gen‑4.5 that predicts the next frame as you move, click, or control a robot arm—turning video generation into an ongoing simulation, not a one‑off render. GWM Worlds lets you turn one scene into a navigable environment for virtual scouting, GWM Avatars syncs faces and bodies to any audio track, and GWM Robotics exposes the same physics‑aware engine as synthetic training data for real robots. Early creator tests show stronger tracking shots and environmental coherence, alongside occasional slapstick glitches that prove the model is still, well, learning to walk.

While GPT‑5.2 chases 50–70%+ scores on ARC‑AGI‑2 and GDPval, Runway is quietly betting that the real power move for filmmakers is treating models as living sets where you can keep shooting forever.