---
title: "Qwen3 Embedding 8B API hosted in the EU"
description: "Run Qwen3 Embedding 8B on inference.eu: multilingual text embeddings for search and RAG, up to 4096 dimensions and 32K tokens, processed in the EU."
canonical: "https://inference.eu/models/qwen3-embedding-8b/"
language: "en"
---

Embeddings · Qwen (Alibaba Cloud)

# Run Qwen3 Embedding 8B in the EU through an OpenAI-compatible embeddings API

Qwen3 Embedding 8B turns text into vectors for semantic search, retrieval-augmented generation, clustering and classification. It covers more than 100 languages, including programming languages, and reads up to 32K tokens per input. On inference.eu you call it through the OpenAI embeddings format, with your documents processed in EU data centres and never used for training.

[Contact sales](https://inference.eu/contact/)[Example request](#example)

1.  018B
2.  02EU data centres
3.  03Apache 2.0

## What Qwen3 Embedding 8B is

Qwen3 Embedding 8B is the largest model in Alibaba's Qwen3 Embedding series, released in June 2025 under the Apache 2.0 licence. It is built on the Qwen3 foundation model, has 8 billion parameters and 36 layers, and produces embeddings of up to 4096 dimensions.

It accepts inputs of up to 32K tokens, so long documents can be embedded without heavy chunking. It supports more than 100 natural and programming languages, which makes it useful for multilingual and cross-lingual retrieval as well as code search. The same series includes reranking models for a second, more precise ranking step.

## What it does well, and where it falls short

The model's strengths are retrieval quality across many languages, long inputs and flexibility. It supports Matryoshka representation learning, so you can request shorter vectors, from 32 up to 4096 dimensions, and trade a little accuracy for much smaller indexes. It is also instruction-aware: a short task description on the query side typically improves results by 1 to 5 percent, according to Qwen.

The trade-off is size. At 8 billion parameters it is far heavier than encoders such as bge-m3 or multilingual-e5-large-instruct, so embedding very large corpora takes more compute. If your documents are short and your index is huge, a smaller model may be the practical choice.

## Calling Qwen3 Embedding 8B through the API

inference.eu uses the OpenAI request format, so any OpenAI SDK, LangChain, LlamaIndex or vector database integration that calls `/v1/embeddings` works unchanged. Set the base URL to the inference.eu API, use your inference.eu key and set the model to `qwen3-embedding-8b`.

Pass a string or a list of strings as `input`. Use the `dimensions` parameter to request shorter vectors that fit your index. For queries, prefix the text with a one-line task, for example `Instruct: Given a web search query, retrieve relevant passages that answer the query` followed by `Query:` and the question. Documents need no instruction. Qwen recommends writing instructions in English, even for other languages.

## Data residency for Qwen3 Embedding 8B

Embedding a knowledge base means sending the whole corpus to the model: contracts, tickets, wikis and customer records. On inference.eu, Qwen3 Embedding 8B runs on infrastructure in the European Union, so this text is processed in EU data centres, is not sent to Alibaba in China and is never used to train any model.

Every customer receives a data processing agreement under Article 28 of the GDPR and our subprocessor list, which helps you document the RAG pipeline in your records of processing. Billing is in euros. The Apache 2.0 licence allows commercial use without user limits.

## Qwen3 Embedding 8B

Maker

Qwen (Alibaba Cloud)

Made in

China

Type

Embeddings

Parameters

8B

Context window

32,768 tokens

Licence

[Apache 2.0](https://huggingface.co/Qwen/Qwen3-Embedding-8B/blob/main/LICENSE)

Released

June 2025

Model ID

qwen3-embedding-8b

Processing

EU data centres

Training on your data

Never

Source

[huggingface.co](https://huggingface.co/Qwen/Qwen3-Embedding-8B)

qwen3-embedding-8b · Python

```
import os
from openai import OpenAI

client = OpenAI(base_url="https://api.inference.eu/v1", api_key=os.environ["INFERENCE_EU_API_KEY"])

result = client.embeddings.create(
    model="qwen3-embedding-8b",
    input=["Where is my data processed?"],
)
print(len(result.data[0].embedding))
```

## What teams use it for

-   1

    ### Multilingual enterprise search

    Index documents in German, French, English and other languages and find them with a query in any of them.

-   2

    ### Retrieval for RAG assistants

    Retrieve the right passages from long policies, manuals or contracts before a chat model writes the answer.

-   3

    ### Code search

    Find functions and snippets across repositories from a plain-language description of what the code does.

-   4

    ### Clustering and classification

    Group support tickets or feedback by topic and route them using vector similarity.

## Qwen3 Embedding 8B: common questions

The questions buyers, developers and legal teams ask us most. Something missing? Ask us directly.

Compliance depends on how you process data. On inference.eu, embedding requests are processed in EU data centres, not used for training and covered by a data processing agreement under Article 28 GDPR with a published subprocessor list. Your documents are not sent to Alibaba or to China.

Qwen3 Embedding 8B reads up to 32K tokens per input. That is enough to embed long sections or whole documents, although shorter chunks often give more precise retrieval in RAG systems. Test chunk sizes on your own data.

Up to 4096. The model supports Matryoshka representation learning, so you can request any size from 32 to 4096 dimensions with the \`dimensions\` parameter. Smaller vectors reduce storage and speed up search at a small cost in accuracy.

Yes. It is released under the Apache 2.0 licence, which allows commercial use, modification and redistribution without fees or user limits. You can use it in paid products and internal systems without a separate agreement with Alibaba.

It is recommended. Qwen reports that a short task instruction on the query side usually improves retrieval by 1 to 5 percent. Documents should be embedded without an instruction. Qwen suggests writing instructions in English, even in multilingual setups.

More than 100 languages, including German, French, Spanish, Italian, Polish and other European languages, as well as many programming languages. It handles cross-lingual retrieval, such as a French query finding a German document.

bge-m3 is a much smaller encoder of about 568 million parameters with 8192-token inputs and dense, sparse and multi-vector output. Qwen3 Embedding 8B is larger, reads longer inputs and offers flexible dimensions. bge-m3 is cheaper to run on huge corpora; test both on your data.

Yes, but vectors from different models are not compatible. Switching means re-embedding your whole corpus and rebuilding the index. The API call itself stays the same: only the model string changes, so the code change is small.

On inference.eu the open weights run in EU data centres, so your documents do not go to Alibaba or to China. An embedding model only returns vectors, not generated text, which limits concerns about content bias. Evaluate retrieval quality on your own data.

Create an OpenAI client with the inference.eu base URL and your inference.eu key, then call \`client.embeddings.create\` with \`model\` set to \`qwen3-embedding-8b\` and your text as \`input\`. Existing vector store integrations work the same way.

[Ask a question](https://inference.eu/contact/)

## Related

[All models](https://inference.eu/models/#directory)

-   [BGE-M3Run bge-m3 on inference.eu: a compact multilingual embedding model for search and RAG, 100+ languages and 8192-token inputs, processed in the EU.Embeddings](https://inference.eu/models/bge-m3/)
-   [multilingual-e5-large-instructRun multilingual-e5-large-instruct on inference.eu: compact embeddings in 100 languages for search, RAG and classification, processed in the EU.Embeddings](https://inference.eu/models/multilingual-e5-large-instruct/)
-   [BGE Reranker v2 M3Run bge-reranker-v2-m3 on inference.eu: a light multilingual reranker that sorts search and RAG results by relevance, processed in the EU.Reranking](https://inference.eu/models/bge-reranker-v2-m3/)

## Move your AI workloads to EU infrastructure

Tell us which models you use and roughly how much traffic you have. We reply within one working day with pricing and next steps.

-   Reply within one working day
-   Pricing for your volume
-   DPA ready to sign