Jina AI embedding model

jina-embeddings-v5-omni-small

Jina v5 Omni Small

Jina v5 Omni Small is Jina AI's multimodal embedding model for text, images, video, audio and PDF in one vector space. it extends jina-embeddings-v5-text-small, returns matching text vectors at the same price, and has a cheaper sibling, omni-nano.

01

price

$0.05/M

per 1M input tokens

embeddings bill input only — there are no output tokens. octomind bills the listed provider price, with no markup, on every plan, including the free one (pay as you go from credits).

02

what real work costs

workloadcost
index 10,000 help articles ~800 tokens each, 8M tokens $0.40
index a 1M-line codebase ~10 tokens a line, 10M tokens $0.50
embed 1M search queries ~20 tokens each, 20M tokens $1.00

at the listed price. re-embedding after a model change costs the same again.

03

specs

provider
Jina AI
max input
about 32k tokens per text
dimensions
1,024 by default; also 32, 64, 128, 256, 512 and 768
storage per 1M vectors
4.1 GB at 1,024 dims · 3.1 GB at 768 dims · 2 GB at 512 dims (float32)
accepts
text, images, video, audio, pdf
multilingual
yes
weights
open — CC BY-NC 4.0
released
May 7, 2026
model id
jina-embeddings-v5-omni-small

integration notes

  • task defaults to text-matching; retrieval needs retrieval.query on queries and retrieval.passage on documents.
  • text over the model's token limit returns an error unless truncate is set to true; truncate defaults to false.
  • a PDF must be sent on its own, not inside a list of inputs.

04

benchmarks

as published by the maker; not measured by octomind.

benchmarkscore
MMTEB (task-level average)source67.0
MTEB (English)source71.7
MIEB (aggregate, image)source58.00
MMEB-Video subsetsource41.20
MAEB (audio)source49.96

05

when to pick it

pick it if

  • you search images, video, audio or PDFs alongside text in one index
  • you already index with v5-text-small; its text vectors match
  • you want 1024-dimension vectors that truncate down to 32

06

where it sits on price

5th cheapest of the 10 embedding models in the catalog.

07

use Jina v5 Omni Small in octomind

octocode, octomind's semantic code search, takes any octohub embedding model by id. one octohub key covers it.

octocode config.toml
[embedding]
code_model = "octohub:jina-embeddings-v5-omni-small"
text_model = "octohub:jina-embeddings-v5-omni-small"

08

compare Jina v5 Omni Small

all model comparisons →

09

similar embeddings

nearest embedding models by price.

view all models →

10

Jina v5 Omni Small faq

how much does jina-embeddings-v5-omni-small cost?

$0.05/M input tokens; embeddings have no output charge. index a 1M-line codebase (~10 tokens a line, 10M tokens) costs about $0.50.

how many dimensions does jina-embeddings-v5-omni-small have?

1,024 by default; also 32, 64, 128, 256, 512 and 768. at the default size, one million float32 vectors take 4.1 GB.

what is the maximum input length of Jina v5 Omni Small?

about 32k tokens per text (≈ 50 pages of prose or 3,000 lines of code). longer documents need chunking before embedding.

how do i use Jina v5 Omni Small with octocode?

set code_model and text_model to "octohub:jina-embeddings-v5-omni-small" in the [embedding] section of octocode's config.toml, then re-index.

use Jina v5 Omni Small in octomind

use it for semantic code search with octocode, or call it through octohub.