Seoul, South Korea - Twelve Labs, a generative artificial intelligence (AI) startup based in South Korea, has unveiled Pegasus-1, a new video language foundation model. This development marks a significant step in the hyperscale large language model market, with Pegasus-1 designed to enhance video summarization and interaction.
According to Yonhap News Agency, Pegasus-1 can summarize long videos into text and engage in conversations about the video content with users. The model reportedly offers 61 percent higher performance compared to the current most advanced video language models. Twelve Labs revealed that Pegasus-1 was trained on over 1 billion image-text pairs and 35 million video-text pairs, a fraction of the 300 million diverse video-text pairs the company has accumulated.
The company sees immediate commercialization potential for Pegasus-1 in various sectors, including sports, media, entertainment, education, and physical security. Founded in 2021, Twelve Labs has already attracted significant investment, securing a combined US$10 million last month from notable investors like U.S. chip design company Nvidia Corp. and U.S. chipmaker Intel Corp.