Platform

Pricing

Solutions

Build

Resources

Company

Select Language

Playground

Talk to Sales

MARENGO 3.5

media & entertainment

Find everything relevant. In every modality.

A multimodal embedding model that processes visual, audio, dialogue, and motion across 36 languages, returning a 512-dimensional vector ready for production search.

A multimodal embedding model that processes visual, audio, dialogue, and motion across 36 languages, returning a 512-dimensional vector ready for production search.

Unbounded

No video/audio duration cap.

36+

Languages for query and retrieval. Native, not translated.

512d

Embedding dimensions 6× smaller than other competitors.

30×

Faster than others at video indexing, while leading on accuracy.

Things only Marengo does.

Other foundation models stop at text-to-clip retrieval on short videos in English. Marengo starts there, then does the work that production systems actually need.

Other foundation models stop at text-to-clip retrieval on short videos in English. Marengo starts there, then does the work that production systems actually need.

512 dimensions. Same understanding.

Marengo encodes video, audio, image, text, and documents into 512-dimensional embeddings, cutting storage 6x versus competitors and speeding searches without losing accuracy.

An image plus a sentence, in one query.

Drop a player photo, add a sentence or two, and Marengo merges them into one embedding. Mix image, text, audio, and documents in a single query.

36 languages, one space.

Search across 36 languages in Marengo's unified vector space. No translation step, no accuracy loss between languages.

Search with fully composable queries.

Use an image query refined by text to find matching video moments.

From signup to first result in 5 minutes.

Same model, prompts, and JSON output. Choose the surface for your team.

Same model, prompts, and JSON output. Choose the surface for your team.

Python

Node.js

import requests
 
# Step 2: Define the API URL and the specific endpoint
API_URL = "https://api.twelvelabs.io/v1.3"
INDEXES_URL = f"{API_URL}/indexes"
 
# Step 3: Create the necessary headers for authentication
headers = {
    "x-api-key": "<YOUR_API_KEY>"
}
 
# Step 4: Prepare the data payload for your API request
INDEX_NAME = "<YOUR_INDEX_NAME>"
data = {
    "models": [
        {
            "model_name": "marengo3.0",
            "model_options": ["visual", "audio"]
        }
    ]
}

Quick Start

Index your first video and run a search in under 5 minutes.

Sample Apps

Production repos: video search, RAG over video, highlight reel generator, compliance scanner.

MCP Server

Connect Claude, Cursor, or any MCP client to your video index.

SDKs & API

Python, Node, REST. Full reference, typed responses, streaming.

Quick Start

Index your first video and run a search in under 5 minutes.

Sample Apps

Production repos: video search, RAG over video, highlight reel generator, compliance scanner.

MCP Server

Connect Claude, Cursor, or any MCP client to your video index.

SDKs & API

Python, Node, REST. Full reference, typed responses, streaming.

Faster, smaller, more accurate.

Marengo was designed against production workloads, not benchmarks, demos, or three-minute clips. Here's what that means in practice.

Marengo was designed against production workloads, not benchmarks, demos, or three-minute clips. Here's what that means in practice.

Composite video-retrieval performance

CAPABILITY

MARENGO 3.5

Gemini Embedding 2

Nova Multimodal embeddings

Embedding dimensions

512

3072 (default; 1536 / 768 via Matryoshka)

3072 (default; 1024 / 384 / 256 via Matryoshka)

Max video length

Unbounded (no duration cap)

120 sec per request

30 sec per segment (chunked)

Multimodal Composite Queries

Yes - video, audio, image, text, and documents composed in one query (16K-token sync endpoint))

No native composite query API

No native composite query API

Sports recognition

Leads by 12–38 pts vs. GE2 on sports/image-query/image+text search

Baseline

Trails Marengo 3.5 by 35–45 pts

Embedding uncertainty signal

Yes - optional uncertainty signal on every embedding

None

None

Start building with TwelveLabs

Bring video intelligence into your production and archive workflows. Deploy, scale, and get more from your content.

Start Building

Talk to Sales

Start building with TwelveLabs

Bring video intelligence into your production and archive workflows. Deploy, scale, and get more from your content.

Start Building

Talk to Sales

Start building with TwelveLabs

Bring video intelligence into your production and archive workflows. Deploy, scale, and get more from your content.

Start Building

Talk to Sales

Desktop

Platform

Platform Overview

Models Overview

Enterprise

Developers

Enterprise Terms of Service

Additional Legal Resources

Capabilities

Media & Entertainment

Sports & Broadcasting

Company

Social

Badge

Tpn

©

2026

TwelveLabs, Inc. All Rights Reserved

Tablet

Platform

Platform Overview

Models Overview

Enterprise

Developers

Enterprise Terms of Service

Additional Legal Resources

Capabilities

Media & Entertainment

Sports & Broadcasting

Company

Social

Badge

Tpn

©

2026

TwelveLabs, Inc. All Rights Reserved

Phone

Platform

Platform Overview

Models Overview

Enterprise

Developers

Enterprise Terms of Service

Additional Legal Resources

Capabilities

Media & Entertainment

Sports & Broadcasting

Company

Social

©

2026

TwelveLabs, Inc. All Rights Reserved

Badge

Tpn