Qwen3-Embedding-8B

Qwen3-Embedding-8B

Turns text into vectors for search, code retrieval, classification and clustering, in over 100 languages. The 8B model in Qwen's embedding family, with adjustable vector size and task instructions.

Made by Qwen on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Qwen3-Embedding-8B is an 8B text embedding model from Qwen, built on Qwen3-8B-Base. It maps text, including code, to vectors that you can use for retrieval, classification, clustering and bitext mining. It supports over 100 languages and a 32k token input.

Who it suits: Developers who build search, retrieval and recommendation features and want one embedding model that handles many languages and programming languages.

Industry | Who uses it | How they use it | Example
Software | Search engineers | Embed documents and queries for semantic search | Find the code snippet closest to a question
Support | Knowledge teams | Match user questions to help articles | Retrieve the right answer page across languages
Data | Analysts | Cluster or classify large piles of text | Group customer feedback by theme

If you want a smaller, cheaper vector model, the series also has 0.6B and 4B versions.

Where to use it:

• Sentence Transformers in Python
• Hugging Face Transformers 4.51.0 or newer
• Windows, macOS or Linux, for downloading the files

Use case | Who | What they get back
Text retrieval | Search engineers | Vectors for matching queries to documents
Code retrieval | Developers | Vectors that link questions to code
Clustering and classification | Analysts | Vectors you can group or label

How to set it up:

• Python with pip
• Transformers 4.51.0 or newer, since older versions fail with a KeyError on qwen3
• Hardware for about 15.1 GB in BF16, or about 3.8 GB at 4-bit by Hugging Face's estimate

1. Install the Hugging Face hub package.
2. Download the model files.
3. Load them with Sentence Transformers or Transformers, as the model card shows.

pip install -U huggingface_hub
hf download Qwen/Qwen3-Embedding-8B

How to use it: Encode your texts with Sentence Transformers and compare the vectors. You can pick an output size from 32 up to 4096 dimensions. Add a task instruction to the query side, and write it in English, since most training instructions were in English.

Tool | What it does
Qwen3-Embedding-8B | Makes embedding vectors up to 4096 dimensions
Qwen3 Reranker models | Rerank retrieved results, and pair with the embedding model
Instruction field | Tailors the vector to a task, language or scenario

Example: You embed a set of help articles once, then embed a user's question with an instruction that says it is a support search. You compare the vectors and return the closest articles.

For newcomers and experts: If you are new, an embedding is a list of numbers that captures a text's meaning, so similar texts get similar numbers. Start by embedding a few sentences and comparing them.

The model supports flexible vector sizes through its MRL support and is instruction aware. The card says that for most tasks, adding an instruction usually improves results by 1 percent to 5 percent, so it recommends writing task-specific instructions.

Text or code goes in
An optional task instruction is added
The model produces a vector
You compare vectors for search or clustering

Good to know:

• Output size is adjustable from 32 to 4096 dimensions
• The series also has Qwen3 Reranker models in the same three sizes
• License: Apache 2.0
• Cost to run: none named, you supply the hardware
• Data it reaches: the text you embed and the files downloaded from Hugging Face
• Not for: Transformers versions older than 4.51.0

Title: Embedding dimension by model size
Qwen3-Embedding-0.6B: 1024 dimensions
Qwen3-Embedding-4B: 2560 dimensions
Qwen3-Embedding-8B: 4096 dimensions

On Hugging Face, 13 November 2026

• Task: feature extraction
• Library: sentence-transformers
• Parameters: 7.6B
• Weights: about 15.1 GB in BF16; about 3.8 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 2,358,928
• Likes: 858
• License: apache-2.0
• Based on: Qwen/Qwen3-8B-Base
• Quantized versions: 45
• Fine-tunes: 49
• Tags: sentence-transformers, safetensors, qwen3, text-generation, transformers, sentence-similarity, feature-extraction, text-embeddings-inference

Links

• Model: https://huggingface.co/Qwen/Qwen3-Embedding-8B
• Files: https://huggingface.co/Qwen/Qwen3-Embedding-8B/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download Qwen/Qwen3-Embedding-8B

Then follow the usage in its model card. Nothing in this kit runs by itself.