TwelveLabs announced the general availability of Marengo 3.0, its most advanced video-understanding model to date, now accessible via Amazon Bedrock and TwelveLabs directly. Marengo 3.0 doesn’t just “watch” video it reads, hears, and comprehends the full rhythm of a scene, effectively linking dialogue to a gesture even minutes later and tracking objects, movement, emotion, and events over time. Built on a multimodal architecture, the model treats video as a living, dynamic system integrating audio, text, visuals and context into a unified embedding space. The release claims a 50% reduction in storage costs and 2× faster indexing performance, with support for four-hour videos and multilingual comprehension across 36 languages.
Also Read: zeb Expands Generative AI Capabilities Through New Multi-Year AWS Partnership
As described in the announcement: “Video represents 90% of digitized data, but that data has been largely unusable because it takes too long for humans to break down, and machines have been incapable of grasping and accounting for everything that happens in video,” said Jae Lee, CEO and co-founder of TwelveLabs. “‘Solving this problem has been our singular obsession. Now, Marengo 3.0 shatters the limits of what is possible. It is an incomparable solution for enterprises and developers.’” The model delivers native video understanding (as opposed to image-derived adaptations), temporal and spatial reasoning including novel sports intelligence features like team, player, jersey number and action tracking and supports composed multimodal queries (image + text) for granular search results. With its API-first design and enterprise-ready deployment, Marengo 3.0 promises to convert vast, underutilized video archives into actionable, searchable, and monetizable assets for industries ranging from media and advertising to security and public sector.
