Google added video Agent to Gemini: cost reduction of up to 66%

Google has online the gemini API on-line agentic video understanding, allowing the model to independently select video clips for inspection. The new model allows for the search of clips along a time axis and selects the image, audio or transcribed text from the question. Google self-measured that, after Gemini 3.7 flash had been opened, token consumption had dropped by a maximum of 88 per cent, analysis costs by a maximum of 66 per cent and accuracy had increased by about 7 per cent. This feature supports the uploading of videos and YouTube videos without additional charges。