Runway shipped Solaris on August 31, and it’s the most radical bet yet on software that gets generated instead of written. It’s a model, not a tool: no HTML, CSS, or JS anywhere. The interface is a 720p video stream rendered frame by frame by Gen-4.5, and your clicks, drags, and typed text become inputs to the next frame.
How the architecture works
Runway split the job in two. An LLM does the reasoning — interpreting what you want, deciding scene transitions — while the world model handles rendering in real time. That split is why Runway claims Solaris beats frontier LLMs (Claude, GPT, Gemini) on structural similarity and information retention when generating new interfaces. In a 250-person study, people preferred Solaris over coded alternatives 61% to 24% on following instructions.
Why it matters
Every “AI builds your app” product so far still writes code underneath. Solaris skips that layer entirely — the first big-lab productization of the world-model-as-software route. The honest caveats: text rendering is shaky, long sessions drift, and there’s no API or pricing. Early access applications only.
You Might Also Like
- Mistrals Next Open Weight Model Opens Early Access in July
- Runway gen 4 5 Just Took the top Spot in ai Video and its not Even Close
- Netflix Void Scores 3 5x Over Runway in Blind Tests Netflixs First Open Source ai Model
- Switchbot Onero h1 a 22 dof Home Robot for 1500 With Full api Access
- X Square Robot Wall b wum World Unified Model Ships Into Chinese Homes may 25 Backed by Xiaomis 276m
Top AI Product 
Leave a comment