Runway Unveils Gen 4.5 AI Model With Realistic Text-to-Video and Benchmark Lead

Runway has announced Gen 4.5, its latest text-to-video AI model that reportedly surpasses competitors like Google's Veo 3 and OpenAI's Sora. The model currently holds the top spot on the independent Video Arena leaderboard, where it secured its ranking through blind human benchmarking that compares models without revealing their origins.

This latest version generates high-definition videos from written prompts, with improvements in understanding physics, human motion, and camera dynamics. Gen 4.5 aims for cinematic realism, with objects and liquids movement with accurate weight, momentum, and better consistency across various visual styles.

While the model is rolling out gradually to users, Runway confirms that generation speed remains as efficient as previous versions. However, the technology still has limitations, including occasional lapses in object permanence and errors in causal reasoning where effects may precede their causes.