Three companies cut video AI token costs using Gemini agentic understanding

Mosaic, Ponder Studio, and Revyl tested Google's agentic video understanding with early-access Gemini Flash models for production video editing, B-roll selection, and mobile app testing.

Mosaic reported a 97% cut in median token usage for its video agent Motion, and Ponder Studio reported a 0.967 F1 score with roughly 72% lower token costs when selecting B-roll.

Revyl reported a 65% accuracy improvement in catching UI bugs and animation stutters, and the capability is available now via the Gemini API for video uploads and YouTube videos.