Runway's Gen-4.5 tops text-to-video leaderboard
Runway has released Gen-4.5, a video generation model that ranks first on the Artificial Analysis Text-to-Video benchmark with 1,247 Elo points, at pricing comparable to its predecessor.
Runway announced Gen-4.5 on December 1, 2025, calling it the top-rated AI video generation model. The company says the model holds the #1 position on the Artificial Analysis Text-to-Video benchmark with 1,247 Elo points, ahead of all other models on that leaderboard.
Runway attributes the gains to "significant advances in both pre-training data efficiency and post-training techniques." It did not give technical details of those methods, beyond saying the model was built on NVIDIA Hopper and Blackwell GPUs.
The company says Gen-4.5 improves motion quality, prompt adherence, temporal consistency and controllability. It highlights physical realism in particular: objects that move with believable weight and momentum, liquids with proper dynamics, and fine details such as hair strands and fabric weave that stay coherent through motion. Runway's demo prompts cover complex multi-object scenes, character expressions, collisions and balancing objects, and styles ranging from photorealistic footage to stop-motion and stylized animation.
According to Runway, Gen-4.5 keeps the speed and efficiency of Gen-4 and is available at comparable pricing on all subscription plans. Existing control modes, including Image to Video, Keyframes and Video to Video, will be brought to Gen-4.5. The announcement does not say when.
The benchmark ranking is the only third-party performance figure Runway cites. The announcement does not include architecture details, model size, or limits on clip length and resolution.