Runway's Gen-4.5 AI Video Model Claims Top Spot, Outperforming Google and OpenAI

Runway says its new text-to-video AI generator has 'unprecedented' accuracy

Runway claims its latest text-to-video model generates even more accurate visuals than its last. In a blog post on Monday, Runway says its Gen-4.5 model can produce "cinematic and highly realistic outputs," potentially making it even more difficult to distinguish between what's real and what's AI. "Gen-4.5 achieves unprecedented physical accuracy and visual precision," Runway's announcement says. It adds that the new AI model is better at adhering to prompts, allowing it to produce detailed scenes without compromising video quality. Runway says that AI-generated objects "move with realistic weight, momentum and force," while liquids "flow with proper dynamics." The Gen-4.5 model is rolling out to all users gradually and will offer the same speed and efficiency as its predecessor, according to Runway. There are still some limitations, though, as the model may experience issues with object permanence and causal reasoning, meaning effects may happen before the cause, such as a door opening before someone uses a handle. Along with Runway, OpenAI is ramping up efforts to make its AI-generated videos look more lifelike. OpenAI highlighted upgrades to physics with the release of its Sora 2 text-to-video model in September, with Sora head Bill Peebles saying, "You can accurately do backflips on top of a paddleboard on a body of water, and all of the fluid dynamics and buoyancy are accurately modeled." Runway says its Gen-4.5 model is better at handling different visual styles, too, allowing it to produce more consistent photorealistic, stylized, and cinematic visuals. The startup claims that photorealistic visuals created with Gen-4.5 can be "indistinguishable from real-world footage with lifelike detail and accuracy."

Runway rolls out new AI video model that beats Google, OpenAI in key benchmark

Gen 4.5 allows users to generate high-definition videos based on written prompts that describe the motion and action they want. Runway said the model is good at understanding physics, human motion, camera movements and cause and effect. The model holds the No. 1 spot on the Video Arena leaderboard, which is maintained by the independent AI benchmarking and analysis company Artificial Analysis. To determine the text-to-video model rankings, people compare two different model outputs and vote for their favorite without knowing which companies are behind them. Google's Veo 3 model holds second place on the leaderboard, and OpenAI's Sora 2 Pro model is in seventh place. "We managed to out-compete trillion-dollar companies with a team of 100 people," Runway CEO Cristóbal Valenzuela told CNBC in an interview. "You can get to frontiers just by being extremely focused and diligent."