ModelCards

voyage-multimodal-3.5

Embedding modelCard summarized

Rich multimodal embedding model that can vectorize interleaved text and visual data

Source: voyage-multimodal-3.5 launch post (no model card published)

Use it via API

$0.12 / 1M tok
Code exampleCompare

Where to call it

  • Voyage AI APIvoyage-multimodal-3.5

Supports

  • Batch API
  • Image input
Full feature support

Good for

  • Retrieval over screenshots, PDFs, slides, tables and figures without OCR
  • Video scene retrieval
  • Mixed text and image search in one index

Watch out for

Overview

Voyage AI · 02Multimodal, front

multimodal-3.5

Rich multimodal embedding model that can vectorize interleaved text and visual data

Typical price
$0.12 / 1M tok
Text input
Dimensions
1024 default
Max input
32,000 tokens per input
Modes
multimodal embeddings, interleaved text + image embeddings
Released
15 Jan 2026

voyage-multimodal-3.5 embeds interleaved text, images and now video into one vector space. Text and images pass through a single transformer encoder rather than separate CLIP-style towers, which Voyage says avoids the modality gap. Video is sent as an ordered sequence of frames. It targets visual document retrieval (screenshots, PDFs, slides, tables) and video retrieval, and Voyage reports text retrieval within 0.29% of voyage-3-large. launch post

voyage-multimodal-3.5
Typical price
$0.12 / 1M tok
Dimensions
1024 default (can be set to 256, 512 or 2048)
Max input
32,000 tokens per input (560 image pixels or 1,120 video pixels = 1 token); images ≤16M pixels and 20 MB; video ≤20 MB
Weights
Closed
Released
15 Jan 2026
  • multimodal embeddings
  • interleaved text + image embeddings
  • video embeddings
  • text embeddings
  • visual document retrieval (screenshots, PDFs, slides)