Jina Embeddings v5
Jina AI's v5 embedding pair: 677M and 239M multilingual models distilled from Qwen3-Embedding-4B, free for non-commercial use only.
In short. Jina Embeddings v5 are two compact multilingual embedding models that turn text into numbers for search. The 677M small model reads up to 32,000 tokens in 119+ languages, and the 239M nano model is lighter still. Jina's own tests show strong scores for their size, but the license is non-commercial, so using them in a paid product needs a separate license.
Summary
Jina AI released jina-embeddings-v5-text on 18 February 2026 and described it in the paper "jina-embeddings-v5-text: Task-Targeted Embedding Distillation". There are two models. The small one has 677M parameters according to its card, is built on Qwen3-0.6B-Base, produces 1,024-dimensional vectors and accepts up to 32,768 tokens in 119+ languages. The nano one has 239M parameters, is built on EuroBERT-210M, produces 768-dimensional vectors and lists an 8,192-token limit in its card. Both support Matryoshka truncation, and Jina describes the vectors as robust under truncation and binary quantization.
The training recipe combines distillation with task-specific contrastive losses: Jina distilled Qwen3-Embedding-4B into the sub-1B models, and the checkpoints ship with four task adapters (retrieval, text matching, clustering and classification), so you pick the one that fits your use. Jina reports averages of 71.7 (small) and 71.0 (nano) on MTEB English v2 and 67.7 and 65.5 on the multilingual MMTEB, and calls the small model the highest-scoring multilingual embedding model under 1B parameters. These are the developer's own results.
The weights are licensed CC BY-NC 4.0, so they are free for research and personal projects but not for commercial products. Elastic completed its acquisition of Jina AI on 9 October 2025 and said enterprise use of the models would be offered through its Elastic Inference Service on Elastic Cloud; self-hosted commercial use would need a separate license from the creator. Jina also publishes v5 omni embedding models and the jina-reranker-v3.5 reranker on Hugging Face under the same non-commercial license. If you need a permissive license in a similar size, look at Qwen3-Embedding, BGE-M3 or Granite Embedding R2.
Good for
- Research, evaluation and personal projects that need strong multilingual embeddings in a small size
- Long inputs of up to 32K tokens with the small model
- Low-storage vector search using short Matryoshka vectors or binary quantization
- Teams already on Elastic, which offers the models in its own inference service
Watch out for
- The CC BY-NC 4.0 license rules out commercial products unless you get a separate license or use Elastic's hosted service
- The nano model's card lists an 8,192-token limit, shorter than the small model's 32,768
- The benchmark scores are developer-reported
- Vectors are not interchangeable between models, so switching means re-indexing
Performance claims come from the developer and have not been independently verified.
How much memory does it need?
Memory figures for each variant of Jina Embeddings v5. Image, video, speech and music pipelines also load a text encoder and a decoder, so the memory needed can be larger than the main checkpoint.
| Variants | Parameters | Download size | Memory needed | Fits on |
|---|---|---|---|---|
| text-nano BF16 | 239M | 0.48 GB | 0.552 GB | Laptop or phone |
| text-small BF16 | 677M | 1.35 GB | 1.55 GB | Laptop or phone |
Download size covers every component needed to run the model, such as the generator, text encoder and decoder. Memory needed is a stated or published measured minimum where one exists; otherwise it is the download size plus about 15%. Larger images, longer clips and higher resolutions need more.
The license in plain words
A Creative Commons license that lets anyone copy, share and adapt the work with credit, but only for non-commercial purposes. Some model cards name it without a version number, and the current version is 4.0.
- Attribution or model name must be shown
See also
Frequently asked questions
- What is Jina Embeddings v5?
- Jina Embeddings v5 are two compact multilingual embedding models that turn text into numbers for search. The 677M small model reads up to 32,000 tokens in 119+ languages, and the 239M nano model is lighter still. Jina's own tests show strong scores for their size, but the license is non-commercial, so using them in a paid product needs a separate license.
- Can I use Jina Embeddings v5 commercially?
- Not without separate permission. Jina Embeddings v5 is released under the CC BY-NC, which allows research and personal use only.
- Is Jina Embeddings v5 open source?
- Jina Embeddings v5 is open-weight, not open source in the strict sense. You can download the model files, but the CC BY-NC adds conditions that the Open Source Initiative definition does not allow.
- Can I run Jina Embeddings v5 on my own computer?
- Yes. The smallest variant, text-nano, needs about 0.552 GB of memory at 4-bit precision, which fits a modern laptop or a mid-range graphics card.
- When was Jina Embeddings v5 released?
- Jina Embeddings v5 was released by Jina AI on February 18, 2026.
- How long a context window does Jina Embeddings v5 have?
- Jina Embeddings v5 supports a context window of up to 32K tokens.
Official links
Sources
Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.
- Jina AI release note: jina-embeddings-v5-text jina.ai
- Elastic Search Labs blog: jina-embeddings-v5-text elastic.co
- jina-embeddings-v5-text-small model card (Hugging Face) huggingface.co
- jina-embeddings-v5-text-nano model card (Hugging Face) huggingface.co
- jina-embeddings-v5-text paper (arXiv 2602.15547) arxiv.org
- Elastic completes acquisition of Jina AI (9 Oct 2025) ir.elastic.co
Release date and facts last verified November 11, 2026.
Related models
| Model | Released | Parameters | Context | License | Smallest variant needs |
|---|---|---|---|---|---|
| Qwen3-Embedding Alibaba Qwen | 600M – 7.57B | 32K | Apache 2.0 (Permissive) | 1.38 GB Laptop or phone | |
| BGE-M3 BAAI | 568M | 8K | MIT (Permissive) | 2.61 GB Laptop or phone | |
| Granite Embedding R2 IBM | 48M – 312M | 32K | Apache 2.0 (Permissive) | 0.115 GB Laptop or phone | |
| EmbeddingGemma 2 Google DeepMind | 270M – 740M | 8K | Apache 2.0 (Permissive) | 0.621 GB Laptop or phone |