Generative AI foundation model developer Twelve Labs said on Tuesday it raised $100 million in Series B funding to implement general intelligence beyond basic video understanding.

The investment was co-led by NEA and Naver Ventures, the report said. Amazon, Radical Ventures, Korea Investment Partners, Index Ventures, Kadriyu Capital and Red Bull Ventures also participated. The funding brings Twelve Labs' total amount raised to more than $207 million.

Twelve Labs CEO and co-founder Lee Jae-sung (이재성) said, "5 years ago, we made a contrarian bet that the basis of machine intelligence is not language but real records in motion." He said language is a byproduct of understanding and video is the data that understanding must answer to.

Twelve Labs has built a frontier foundation model capable of understanding video. It focused on implementing a multimodal model that understands video natively, rather than a large language model that simply processes video.

Its flagship products are the Marengo model family and Pegasus 1.5.

Marengo 3.0, launched late last year, embeds video, audio, text and components to fit real-world environments, the company said. It converts various content types into machine-readable data structures such as vector databases, enabling AI models to understand and retrieve information at scale.

Pegasus works with Marengo and converts video into structured data. It identifies scene boundaries, entities, time segments and events, enabling LLMs to reason based on visual information. This is similar to how LLMs summarise vast documents or images into markup language to make them easier to understand.

Current LLMs cannot process video in one go. They must split video into multiple screenshots and then reason from them. Twelve Labs said it has built reasoning capabilities that natively understand temporal flow by maintaining memory that persists and does not disappear with each query.

Based on this, Twelve Labs aims to lead a new video recognition paradigm in which machines can analyse, search and use video in practical work.

Twelve Labs will also strengthen cooperation with Amazon Web Services. Following the investment, it signed a multi-year contract to optimise video inference workloads for AWS Trainium chips. It also plans to launch new frontier models first on AWS.