The artificial intelligence company Runway announced its new video generation model, Gen-4.5, on December 1, 2025. According to the independent benchmarking firm Artificial Analysis, the new model has achieved the top rank in its text-to-video leaderboard, placing it ahead of models from larger competitors like Google and OpenAI.
Gen-4.5 allows users to create high-definition videos by typing text descriptions. Runway states that the model offers significant improvements in motion quality, visual detail, and the ability to follow a user’s prompt accurately. In a statement to CNBC, Runway CEO Cristóbal Valenzuela noted the achievement, saying, “We managed to out-compete trillion-dollar companies with a team of 100 people.” He added that the model was internally codenamed “David” in a reference to the biblical story of David and Goliath.
According to the Artificial Analysis benchmark dated November 30, 2025, Google’s Veo 3 model holds the second-place position, while OpenAI’s Sora 2 Pro is ranked seventh. The rankings are determined by human evaluators who compare videos from two different models without knowing their origin and vote for their preference.
Capabilities and Limitations
Runway claims that Gen-4.5 can generate complex scenes with realistic physics and create expressive, lifelike characters. The company also highlights the model’s ability to maintain a consistent visual style, whether photorealistic or animated. The model was developed in collaboration with NVIDIA and runs on its GPU hardware.
Despite its advancements, Runway is transparent about the model’s current limitations, which are common in today’s video generation technology. These include:
- Causal reasoning issues, where an effect might appear before its cause.
- Problems with object permanence, where objects can appear or disappear unexpectedly between frames.
- A “success bias,” where actions are more likely to succeed than they would in reality.
Runway has stated that Gen-4.5 is rolling out gradually and will be available to all customers by the end of the week through its platform and programming interface (API).
