ModelCards

jina-embeddings-v5-omni-small

Embedding modelOpen weightsCard summarized

Multimodal embeddings for text, image, audio, video, and PDF

Source: jina-embeddings-v5-omni: Geometry-preserving Embeddings via Locked Aligned Towers (GELATO) (technical report, arXiv 2605.08384)

Use it via API

$0.05 / 1M tok
Code exampleCompare

Where to call it

  • Jina Embeddings APIjina-embeddings-v5-omni-small
  • Self-host (weights)jinaai/jina-embeddings-v5-omni-small
  • Google Vertex AIjina-embeddings-v5-omni-small

Supports

  • Image input
  • PDF input
Full feature support

Good for

  • Cross-modal search: a query in one modality retrieves any other
  • Visual document and PDF retrieval
  • Adding images, audio or video to an existing v5-text index without re-embedding

Watch out for

Overview

Jina AI · 02Omni, front

v5 Omni Small

Multimodal embeddings for text, image, audio, video, and PDF

Typical price
$0.05 / 1M tok
Text input, Jina API (1B-token pack)
Dimensions
1024
Max input
32,768 tokens
Modes
text embeddings, image embeddings
Released
7 May 2026

jina-embeddings-v5-omni-small embeds text, images, audio, video and PDFs into one 1024-dimension space. It keeps the jina-embeddings-v5-text-small backbone and frozen vision and audio encoders, and trains only small projectors and delimiter tokens (about 0.35% of weights). Text vectors are bit-identical to v5-text-small, so existing text indexes can gain other modalities without re-embedding. tech report abstract; API FAQ

jina-embeddings-v5-omni-small
Typical price
$0.05 / 1M tok
Dimensions
1024 (Matryoshka: 32, 64, 128, 256, 512, 768, 1024; API accepts 1–1024)
Max input
32,768 tokens; images up to 5 MB, PDFs up to 8 MB (one per request); video sampled to 32 frames
Languages
93 languages supported (trained on 32)
Weights
Open · CC BY-NC 4.0 (commercial self-hosting needs a Jina On-Prem license from Elastic)
Released
7 May 2026
  • text embeddings
  • image embeddings
  • audio embeddings
  • video embeddings
  • PDF / visual document embeddings
  • multimodal embeddings (fused content groups)
  • retrieval, text matching, clustering, classification