Marengo 3.5 is an embedding model for comprehensive video understanding. It analyzes video, audio, image, and document files, and combines their visual, audio, and text information. This provides a holistic understanding similar to human comprehension.

Marengo 3.5 produces 512-dimensional embeddings for faster processing and reduced storage.

Notes

The /search endpoint does not support Marengo 3.5. To search your content with the platform, enable Marengo 3.0 for your index. To search Marengo 3.5 embeddings, query them in your own system.
Marengo 3.5 embeddings are not compatible with Marengo 3.0 embeddings. To move to Marengo 3.5, regenerate your embeddings with Marengo 3.5. For step-by-step instructions, see the Migrate from Marengo 3.0 to Marengo 3.5 guide.

New in Marengo 3.5

Compared to Marengo 3.0, Marengo 3.5 adds the following capabilities:

Capability	Marengo 3.5	Marengo 3.0
Document inputs	PDF, with one embedding per page	Not supported
Video and audio duration	No limit	Up to 4 hours
Composed queries across modalities	Combine text with images, video, or audio	Combine text with images
Context window	16,000 tokens	500 tokens
Audio track	Unified encoder for speech, music, and non-dialog audio	Separate transcription embedding
Embedding uncertainty	Per-dimension uncertainty vector; higher value shows lower confidence	Not available
Time-based metadata fusion	Supported	Not available

Context window

Marengo 3.5 uses a context window of 16,000 tokens. The context window is the maximum number of tokens a single request can use. This limit applies to queries. It does not limit the length of the content you embed.

The following inputs count toward the context window:

Query text
Image content
Video content
Audio content
Document content

Use cases

Create embeddings from video, audio, images, and documents. Use these embeddings for similarity search, content classification, clustering, recommendations, or Retrieval-Augmented Generation (RAG).

The platform provides two ways to create embeddings. To embed a query, the platform processes your request synchronously and returns the embedding in the response. To embed content at scale, the platform processes your media files asynchronously, one file per request. For details, see the Embed a query and Embed content at scale guides.

Input requirements

The specifications on this page reflect the maximum capabilities of the model. Your actual requirements depend on the upload method and operation you choose. For details about the available upload methods and the corresponding limits, see the Upload and processing methods page.

Video file requirements

Duration: 4 sec or longer
File size: No model limit. Your upload method sets the effective limit.
Resolution: 360x360 to 5184x2160
Aspect ratio: Between 1:1 and 1:2.4, or between 2.4:1 and 1:1. For example, you can use 1:1, 4:3, 4:5, 5:4, 16:9, 9:16, or 17:9.
Formats: FFmpeg supported

Notes

If you upload files using publicly accessible URLs, use direct links to raw video files that play without user interaction or custom video players (example: https://example.com/videos/sample-video.mp4). Video hosting platforms and cloud storage sharing links are not supported.
For videos in other formats or if you require different options, contact us at support@twelvelabs.io.

Image file requirements

Formats: JPEG, PNG
Minimum size: 128x128 pixels
File size: ≤ 32 MB

Audio file requirements

Formats: WAV (uncompressed), MP3 (lossy), and FLAC (lossless)
Duration: No limit
File size: No model limit. Your upload method sets the effective limit.

Document file requirements

Format: PDF

Supported languages

Marengo 3.5 supports the following languages:

Arabic, Bengali, Chinese (Simplified), Croatian, Cusco, Czech, Danish, Dutch, English, Farsi, Filipino, Finnish, French, German, Greek, Hebrew, Hindi, Hungarian, Indonesian, Italian, Japanese, Korean, Maori, Norwegian, Polish, Portuguese, Romanian, Russian, Spanish, Swahili, Swedish, Telugu, Thai, Turkish, Ukrainian, and Vietnamese.

Support

For support or feedback regarding Marengo, contact support@twelvelabs.io.