embeddings.app
← All models
Jina AI · Model guide

Jina Embeddings 3

Awaiting activation

An established multilingual embedding model with behavior tailored to the task.

What it does

Jina Embeddings 3 supports multilingual text representation with task-specific adapters. The original model distinguishes retrieval queries, retrieval passages and other tasks such as text matching or clustering, rather than applying exactly the same behavior to every input.

embeddings.app exposes its retrieval query and passage paths. This model is retained for applications that already use its vectors, with a default 1,024-dimensional output and support for shorter vector representations. Its source context is 8,192 tokens.

Where it helps

01

Existing search indexes

Keep questions consistent with a collection already indexed using Jina 3.

02

Multilingual support

Retrieve relevant passages for users asking in different languages.

03

Knowledge retrieval

Connect questions with useful sections of reference material.

When to choose it

Consider it when you already use Jina 3 or need a stable baseline for migration. For a new text retrieval project, evaluate Jina 5 Small and Nano as well.

Awaiting provider activation and verified account pricing. This service selects retrieval behavior through Query or Document; it does not expose every task adapter from the original model.

Official sources

Paraphrased from the provider’s documentation and original model information. Selection advice is our editorial guidance.

Jina AI · Embeddings and model guidance ↗

Reviewed October 3, 2026