The GDELT Project

Reaching Nearly 3M Global TV News Broadcasts Totaling 6B Seconds & 109 Billion Tokens Translated By Gemini For $74.6K

Last month we announced the incredible milestone that we had completed machine translation into English of the entire quarter-century Internet Archive Television News Archive using Gemini. In the weeks since, the incredible new capabilities unlocked by Gemini have allowed us for the first time to process an additional 600K broadcasts that we've never before been able to incorporate into our analyses due to various technical issues. With this new set of broadcasts, we have now translated 3 million broadcasts totaling 6 billion seconds of airtime (100M minutes / 1.7M hours) spanning 9.5 million words over 62 billion characters. In all, the entire translation process consumed 109 billion input+output tokens using Gemini 2.5 Flash Non-Thinking and cost $74,634.  Only the public enterprise Vertex AI Gemini API was used and no data was used to train or tune any model.