ModelCards

Marengo 3.0

Embedding modelCard summarized

Marengo 3.0 is an embedding model. It analyzes visual, audio, and text information.

Source: Marengo 3.0 launch post (no model card published)

Use it via API

$0.07 / 1K req
Code exampleCompare

Where to call it

  • TwelveLabs APImarengo3.0
  • Amazon Bedrocktwelvelabs.marengo-embed-3-0-v1:0

Supports

  • Image input
Full feature support

Good for

  • Any-to-any search with text, images, video clips or audio as the query
  • Finding logos, on-screen text and small objects (down to about 10% of the frame)
  • Motion search, object counting and cinematography terms (zoom, pan, tracking shot)

Watch out for

Overview

TwelveLabs · 01Marengo, front

Marengo 3.0

Marengo 3.0 is an embedding model. It analyzes visual, audio, and text information.

Typical price
$0.07 / 1K req
Text embedding
Dimensions
512
Max input
4 h of video or audio
Modes
video embeddings, audio embeddings
Released
17 Nov 2025

Marengo 3.0 is the TwelveLabs embedding model behind its Embed and Search APIs. It encodes video, audio, images and text into one 512-dimensional space for any-to-any retrieval. It added 36-language queries, composed text-plus-image search, longer text (500 tokens), up to 4 hours of video or audio, and better sports, cinematography and small-object recognition than Marengo 2.7. Marengo 3.5 (31 Aug 2026) supersedes it for embeddings, but platform search still uses 3.0. Launch post (dated 30 Nov 2025); release notes (12 Sep, 14 Oct, 29 Oct, 17 Nov 2025)

marengo3.0
Typical price
$0.07 / 1K req
Dimensions
512
Max input
4 h of video or audio (async), ≤ 4 GB; 10 min on the sync endpoint; 500 text tokens
Languages
36 languages, including English
Weights
Closed
Released
17 Nov 2025
  • video embeddings
  • audio embeddings
  • image embeddings
  • text embeddings
  • semantic video search (any-to-video)
  • composed text + image search
  • entity search
  • transcription (spoken-word) search