Marengo 3.5
Awaiting activationFind meaning in motion, speech, music and visual scenes.
What it does
Marengo 3.5 is TwelveLabs’ current multimodal embedding model. It combines video, audio, images and documents in one space, with an audio encoder covering speech, music and non-dialog sound. It supports 128, 256 or 512 dimensions.
The official synchronous query contract supports composed text and media, up to ten media sources and short audio/video clips. Large files use a separate asynchronous content-at-scale workflow. This is especially relevant for media archives, training footage and recorded business content. Its 3.5 vectors are not compatible with Marengo 3.0; the provider’s hosted search endpoint currently uses 3.0, so 3.5 vectors must be searched in your own system.
Content types
text · image · audio · video · PDF documents
Video and sound retrieval that combines motion, speech, music and documents. Integration under review.
Integration under review. No callable media endpoint is offered for this model yet.
Where it helps
Media archives
Find a scene using what happens or what is heard.
Training and recorded content
Discover relevant moments across instructional footage and recordings.
Sound-aware discovery
Retrieve content using speech, music and environmental sound.
When to choose it
A strong evaluation candidate when motion and sound matter together. Compare it with general omni models on real clips. Integration is under review; no callable model endpoint is offered here yet.
Integration under review. Always use a compatible model, task and dimension for indexed content and queries.
Official sources
Paraphrased from the provider’s documentation and original model information. Selection advice is our editorial guidance.
TwelveLabs Marengo 3.5 · Official model documentation ↗Reviewed October 4, 2026