Gemini's Flash models now dynamically scan videos across frames, audio, and transcripts, slashing token use by 88% and boosting accuracy by 7%.

Full article content could not be extracted automatically. Read the original below.