Runway has introduced Gen-4.5, a video generation model that reportedly takes the top spot on the Artificial Analysis leaderboard with improved physical accuracy and prompt adherence. The update refines the previous architecture to deliver higher fidelity without increasing inference costs.
Key Details
Physics Upgrade: The model specifically targets motion dynamics, with better handling of weight, momentum, and complex interactions like fluid flow or cloth movement.
Benchmark Leader: Gen-4.5 achieved an Elo score of 1,247, which Runway notes places it above competitors like Google Veo and OpenAI’s Sora 2 Pro in blind testing.
Same Price, Better Specs: The model runs on Nvidia’s Hopper and Blackwell GPUs, maintaining the generation speed of Gen-4 while improving prompt adherence and visual consistency.
Known Issues: In a move toward transparency, Runway explicitly listed current limitations, including occasional failures in object permanence and causal reasoning (like effects appearing before causes).
While raw generation quality is up, Runway's decision to highlight specific technical limitations suggests a shift toward managing professional expectations rather than just fueling the hype cycle.
