Qwen3 Embedding 8B
Neural Network
Qwen3 Embedding 8B is a Qwen model for text vector representations: Qwen3 embedding API for semantic search and RAG.
Specifications
- Input
- text
- Output
- vectors (embeddings)
- Context size
- 40 960 tokens
- Vector dimensions
- 4 096
Prompt cost
(per 1M tokens)
How it works Qwen3 Embedding 8B?
Qwen3 Embedding 8B is a model from the Qwen family designed for building text vector representations and ranking tasks; it inherits the multilingualism and long-text handling capabilities characteristic of the Qwen3 line. It connects via the unified OpenAI-compatible BotHub API: access from Russia without VPN or foreign cards, pay-as-you-go in rubles for tokens, and switching to any of the 250+ available models requires no integration rewrite. Vectors are stored in Qdrant or pgvector. It is used for semantic search over knowledge bases, RAG over documentation, clustering and deduplication of requests, reranking of search results, and matching similar products or vacancies.Frequently asked questions about Qwen3 Embedding 8B
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
The model converts text into numerical vectors and is suitable for semantic search, RAG, ranking, deduplication, and document clustering. The context window is 32,000 tokens, so you can vectorize entire articles, tickets, or large fragments of a knowledge base without breaking them into small pieces.
The output is not text or an image, but an embedding — an array of floating-point numbers. Vectors are compared by cosine similarity and stored in a vector database, such as Qdrant or pgvector. It is most convenient to check the exact dimension directly in the API response.
Reasoning before answering is not noted in our data, and for an embedding model, this is not the main task: it does not conduct a dialogue or build chains of reasoning, but encodes the meaning of the text into a vector. You implement the search and ranking logic on your side.
Function calling and structured JSON output are not noted in our data: the model returns a vector representation of the text, not instructions for tools. If you need function calling and strict schemas, connect a generative model via the unified OpenAI-compatible BotHub API, and leave embeddings for search.
Acceptance of anything other than text is not noted in our data, so expect text input. For multimodal scenarios, keep a suitable model nearby: in BotHub, switching happens in one window and via one API, so you won't have to rewrite the integration.
We do not have data on languages, so we will not promise specific quality in Russian. Check it on your own corpus: collect a few dozen query-document pairs and compare the vector similarity. You can run such a test from Russia, without a VPN or foreign card, with payment in rubles.