Jina AIJina Embeddings v4

Jina Embeddings v4

Jina Embeddings v4 encodes text, images and visually rich documents. It supports dense search and multi-vector matching, with dedicated adapters for retrieval, text matching and code.

How the model works

Jina adapts Qwen2.5-VL-3B-Instruct and uses mean pooling for its dense output. The default dense vector has 2,048 dimensions, with trained reductions to 128. Multi-vector retrieval instead produces 128-dimensional token representations. The context limit is 32,768 tokens. Tables, figures and page layout can therefore contribute directly to document retrieval.Jina Embeddings v4 model details

Use and access

The model card explicitly corrects an early license error: Qwen Research License applies, not the initially listed CC BY-NC 4.0. Merged vLLM task packages and GGUF files remain distributions of this model. The API and downloaded model are separate access routes. Do not infer commercial weight rights from the availability of hosted inference or a converted package.Jina Embeddings v4 model detailsJina Embeddings v4 technical overview

Specifications & access

Jina Embeddings v4 specifications
DetailValue and conditions
Launch date
AccessAPI, Downloadable
Model identifiersjina-embeddings-v4jinaai/jina-embeddings-v4jina-embeddings-v4-vllm-retrievaljina-embeddings-v4-vllm-text-matchingjina-embeddings-v4-vllm-code
Context window32,768 tokens

Maximum input sequence in the model card; serving limits can be lower.

Jina Embeddings v4 model details
ParametersApproximately 3.8 billionJina Embeddings v4 model details
Embedding dimensions2,048 dense; 128 per token for multi-vector outputJina Embeddings v4 model details
Weight licenseQwen Research License; corrected from an earlier CC BY-NC labelJina Embeddings v4 model details
Base modelQwen2.5-VL-3B-InstructJina Embeddings v4 model details

Related models

  • Jina Embeddings v5 Omni Small

    Jina AI · Jina Embeddings v5 Omni

    A shared search encoder for text, images, audio, video and document files.

    Launch date API · Downloadable
  • Jina ColBERT v2

    Jina AI · Jina ColBERT

    A multilingual retriever that keeps separate vectors for document tokens.

    Launch date API · Downloadable