Runway’s new Gen-4.5 text-to-video model that beats OpenAI’s Sora 2, Google’s Veo 3

The new Gen-4.5 model is capable of producing video outputs that capture realistic and fluid movements with accuracy.

The Gen-4.5 is capable of handling a wide range of styles, from cinematic to photorealistic animation. (Image: Runway)

New York-based AI company Runway has introduced its latest video generation model, Gen-4.5, which it claims has surpassed the performance of rival text-to-video models from OpenAI and Google.

“Runway Gen-4.5 is the world’s top-rated video model, offering unprecedented visual fidelity and creative control. It produces cinematic and highly realistic outputs while providing limitless creative freedom and precise control over every aspect of generation,” the company said in its official blog announcing the launch.

The company came into the spotlight after launching its Gen-1 model in 2023. Runway claims that Gen-4.5 pushes the frontier of video generation to new heights. The new model represents significant advances in both pre-training data efficiency and post-training techniques.

Also Read | AWS unveils new AI models, Trainium3 servers, Frontier Agents, and more at re:Invent 2025

Further, the company said that the new model sets new standards for temporal consistency and controllable action generation across diverse generation modes. In the benchmark named Elo, Gen-4.5 secured a score of 1247, which puts it marginally ahead of Google’s Veo 3 with a score of 1226, and OpenAI’s Sora 2 Pro with 1206 points.

Reportedly, the model produces excellent quality without compromising on the performance and maintains the speed and efficiency of Gen-4. One of the notable aspects of the model is that it delivers ‘unprecedented physical accuracy’ and visual precision. This means it can generate videos with realistically moving objects, accurate dynamics for flowing liquid, and finer details like hair strands and material weaves that stay consistent across time and motion.

When it comes to aesthetics, the Gen-4.5 is capable of handling a wide range of styles, from cinematic to photorealistic animation. The Gen-4.5 model has been built in collaboration with Nvidia, and its training and inference have been run on Blackwell and Hopper GPUs. Despite the advancements, reportedly the model has the usual issues. Runway claimed that the Gen-4.5 struggles with causality and object permanence.

Besides, the model is also prone to ‘success bias,’ meaning it may produce incorrect outcomes, for example, a poorly aimed arrow hitting the target when it is supposed to fail. The company, however, feels that these issues are crucial to building dependable world models, and it plans to keep iterating. The model is available across all subscription plans, making it accessible to creators and companies at scale.