Google DeepMind just dropped its first Gemini Omni model, and it’s already outperforming nearly every competitor in one of the most closely watched AI benchmarks for video editing. Gemini Omni Flash landed the number two spot on the Video Edit Arena leaderboard with a score of 1347, roughly 39 points ahead of the third-place finisher.
For a model that was announced barely weeks ago at Google I/O on May 19, 2026, that’s a remarkably strong debut. The only model sitting above it is ByteDance’s dreamina-seedance-2.0-720p, which leads by approximately 30 points.
What Gemini Omni Flash actually does
Gemini Omni Flash handles text, images, audio, and video as inputs, then generates or edits video content based on conversational instructions. The “Omni” branding signals Google DeepMind’s push toward true multimodal convergence, where a single model processes and generates across multiple content types simultaneously rather than being siloed into one task.
The model isn’t just showing up in the Video Edit category either. Early performance data indicates strong rankings in Text-to-Video and Image-to-Video categories as well. Its strongest showing appears to be in instruction following and editing, based on human preference evaluations where real users judge model outputs.
