Runway`s new AI video model beats Google and OpenAI benchmark

Runway on Monday introduced Gen 4.5, a new artificial intelligence video model that produces high-definition clips from written prompts. The company said the system interprets descriptions of motion and action and can handle physics, human movement, camera behavior and cause-and-effect sequences.

Gen 4.5 is ranked first on the Video Arena leaderboard, which is run by the independent benchmarking firm Artificial Analysis. The leaderboard compares text-to-video models by asking voters to choose between anonymous outputs from different systems. Google’s Veo 3 model is in second place, and OpenAI’s Sora 2 Pro model is in seventh.

Runway CEO Cristóbal Valenzuela said the company’s small team outperformed far larger rivals in the competition. He told CNBC that a focus on research and consistency helped the firm reach the current milestone. He added that the project carried the internal codename “David,” a reference to the biblical story of David and Goliath.

Runway was founded in 2018. The startup develops video and world models based on video and observational data, aiming to improve the representation of physical interactions in digital environments. Its customer base includes media outlets, film studios, brands, designers, creative professionals and students.

The company’s valuation has reached $3.55 billion, according to PitchBook. Its investors include General Atlantic, Baillie Gifford, Nvidia and Salesforce Ventures.

Valenzuela said Gen 4.5 reflects a shift toward efficient and targeted research in the AI sector. He said the company wants to help ensure that the industry is not dominated by a small number of major players.

Gen 4.5 is being released in stages and will be accessible to all customers by the end of the week. Runway said the model will be available through its platform, its application programming interface and selected partners. Valenzuela added that the company plans several major product releases following Gen 4.5.