Qwen3 Embedding 8B API

Convert text to vectors for search, ranking and clustering in over 100 languages, from $0.009.

70.58 MTEB multilingual
7.6B parameters
33K context
Apache 2.0 license

from openai import OpenAI

client = OpenAI(
    base_url="https://api.modelsheep.com",
    api_key=API_KEY_MODELSHEEP,
)

response = client.embeddings.create(
    model="Qwen/Qwen3-Embedding-8B",
    input="The quick brown fox jumps over the lazy dog.",
)

print(response.data[0].embedding)

How to use the Qwen3 Embedding 8B API

1
Create a ModelSheep account and copy an API key from the dashboard.
2
Point the OpenAI SDK at the base url https://api.modelsheep.com and set the model to Qwen/Qwen3-Embedding-8B.
3
Call /embeddings with your text, put Instruct: {task}\nQuery: {query} in front of a query for higher accuracy, and read back one vector of 4096 numbers per input.

What Qwen3 Embedding 8B is

Qwen3 Embedding 8B converts text to one vector per input. Qwen built it on the Qwen3 foundation model for retrieval, ranking and classification. Qwen trained it on over 100 languages, programming languages among them. It is the largest of the three Qwen3 embedding models. It returns the widest vector, and it scores highest of the three on the MTEB multilingual v2 leaderboard, at 70.58. Choose it over the 4B and the 0.6B when retrieval accuracy decides the product. The smaller two write narrower vectors, so an index built on them costs less memory.

What it does well

Ranks a document store by meaning across languages from one query vector.
Finds the function that matches a plain English description of the code.
Groups support tickets, products or news items by topic without labels.

What it is not for

It returns vectors only. It writes no text, answers no question and calls no
It is the largest of the three and returns the widest vector. On a large

Calling it

Post to /embeddings with the model string Qwen/Qwen3-Embedding-8B. The input field takes one string or an array of strings. An array returns the vectors in the same order, and one call costs less than one call per string. Add dimensions to shorten the vector and save index memory. Embed the queries and the documents at the same width, or the similarity scores mean nothing. A query gains accuracy from an instruction line in front of it, in the form Instruct: {task}\nQuery: {query}. Documents need no instruction.

Qwen/Qwen3-Embedding-8B /embeddings

SDKs and languages

Qwen3 Embedding 8B answers on an HTTP endpoint, so any language that can post JSON reaches it. These are the ones the box above writes out.

OpenAI SDK: PythonTypeScriptGoJavaC#PHPRust
HTTP: cURLPythonTypeScriptGoRustJavaC#PHP

Specifications

Author	Qwen
Category	embedding
Accepts	text
Returns	vector
Total parameters	7.6B
Active parameters	dense, all of them
Score	70.58 on MTEB multilingual v2
Languages	100+One claim for the whole Qwen3 Embedding series. The total counts programming languages, so the natural language count is smaller and unstated.
Precision served	FP8
Context	33K tokens
Vector	4096 dimensions
License	Apache 2.0

Price

Input

$0.009

per million tokens

Output

n/a

this model returns a vector

Against OpenRouter

10%

below the cheapest of 2 providers

The cheapest OpenRouter price for this model was $0.010 in and n/a out, with a median of $0.010 and n/a across the comparable providers. Read on 2026-09-17.

License

Apache 2.0. It permits commercial use, changes and redistribution. It requires you to keep the notice. No vendor terms sit on top of it.

Parameters: 7.953B
Context: 33K
Vector: 4096
Input per 1M: $0.031
Output per 1M: n/a
License: OpenMDW 1.1

Qwen3 Embedding 4BQwen/Qwen3-Embedding-4B

Parameters: 4B
Context: 33K
Vector: 2560
Input per 1M: $0.009
Output per 1M: n/a
License: Apache 2.0

Qwen3 Embedding 0.6BQwen/Qwen3-Embedding-0.6B

Parameters: 0.6B
Context: 33K
Vector: 1024
Input per 1M: $0.004
Output per 1M: n/a
License: Apache 2.0

Start using Qwen3 Embedding 8B

Point the OpenAI SDK at api.modelsheep.com. Your prompts stay in Germany, and nothing is retained.

Get API key Read the quickstart