Qwen3-Embedding-4B is the 4B text-embedding edition in Qwen’s embedding and reranking family. Built on the dense Qwen3 foundation model, it converts a text segment into a numerical representation of its meaning. Qwen identifies text retrieval, code retrieval, classification, clustering and bitext mining as application areas. It is useful when an application needs to find related passages, group similar records or compare text across languages. The model’s output is an embedding vector, not a conversational response or a rewritten document.
The exact edition has a 32K-token sequence length and an embedding dimension of up to 2,560. It supports custom output dimensions from 32 to 2,560 through Matryoshka Representation Learning. The chosen vector length affects how the representation is stored and compared, so keep it consistent between query and document vectors in one collection. Qwen3-Embedding-8B has a different model size and a 4,096-dimension maximum; neither that output size nor the 8B model’s identity transfers to this 4B edition. An existing index should retain its own consistent representation choices.
The producer describes support for more than 100 languages, including programming languages, with multilingual, cross-language and code retrieval. An illustrative help-centre search can compare a user’s question with passages written in several languages, then return the original passages that match the topic. A code search can compare a natural-language description with relevant functions and their documentation. Preserve meaningful headings, domain terms and identifiers when preparing text, and keep each represented passage connected to its source so the application can show the matching material.
Qwen3-Embedding-4B is instruction-aware. Qwen recommends giving each query a one-sentence description of the task, tailored to the intended retrieval relationship, and recommends English instructions for multilingual scenarios. The direct example places Instruct: before the task description and Query: before the query text. Its retrieval documents are encoded without that query instruction. A practical original instruction can specify finding documentation passages that explain a reported error or finding source-code excerpts that implement the described behaviour. This directs representation towards the application’s question rather than asking the model to write an answer.
The official examples encode query and document texts, compare their embeddings using cosine similarity and demonstrate pooling the final token’s hidden state followed by normalisation. In that lower-level example, the input is truncated at a chosen 8,192-token setting; the documented model sequence length remains 32K. For long material, decide which passages to represent and ensure that truncation does not silently discard the part the query needs. Store vectors with stable source references, compare the prepared query with the collection and return the associated passages in similarity order.
Qwen’s embedding model uses a dual-encoder approach, whereas its separate reranking models use a cross-encoder to evaluate query-document pairs. The release describes using those modules together: embeddings find initial candidates, and a reranker can then refine their order. Qwen3-Embedding-4B itself supplies the single-text representation. Its vector is not a reranker score, a verified statement or a generated answer. If the final workflow answers questions from retrieved material, the selected source passages remain the evidence passed to that later stage.
", "aboutText": "Qwen3-Embedding-4B is the 4B text-embedding edition in Qwen’s embedding and reranking family. Built on the dense Qwen3 foundation model, it converts a text segment into a numerical representation of its meaning. Qwen identifies text retrieval, code retrieval, classification, clustering and bitext mining as application areas. It is useful when an application needs to find related passages, group similar records or compare text across languages. The model’s output is an embedding vector, not a conversational response or a rewritten document.\n\nThe exact edition has a 32K-token sequence length and an embedding dimension of up to 2,560. It supports custom output dimensions from 32 to 2,560 through Matryoshka Representation Learning. The chosen vector length affects how the representation is stored and compared, so keep it consistent between query and document vectors in one collection. Qwen3-Embedding-8B has a different model size and a 4,096-dimension maximum; neither that output size nor the 8B model’s identity transfers to this 4B edition. An existing index should retain its own consistent representation choices.\n\nThe producer describes support for more than 100 languages, including programming languages, with multilingual, cross-language and code retrieval. An illustrative help-centre search can compare a user’s question with passages written in several languages, then return the original passages that match the topic. A code search can compare a natural-language description with relevant functions and their documentation. Preserve meaningful headings, domain terms and identifiers when preparing text, and keep each represented passage connected to its source so the application can show the matching material.\n\nQwen3-Embedding-4B is instruction-aware. Qwen recommends giving each query a one-sentence description of the task, tailored to the intended retrieval relationship, and recommends English instructions for multilingual scenarios. The direct example places Instruct: before the task description and Query: before the query text. Its retrieval documents are encoded without that query instruction. A practical original instruction can specify finding documentation passages that explain a reported error or finding source-code excerpts that implement the described behaviour. This directs representation towards the application’s question rather than asking the model to write an answer.\n\nThe official examples encode query and document texts, compare their embeddings using cosine similarity and demonstrate pooling the final token’s hidden state followed by normalisation. In that lower-level example, the input is truncated at a chosen 8,192-token setting; the documented model sequence length remains 32K. For long material, decide which passages to represent and ensure that truncation does not silently discard the part the query needs. Store vectors with stable source references, compare the prepared query with the collection and return the associated passages in similarity order.\n\nQwen’s embedding model uses a dual-encoder approach, whereas its separate reranking models use a cross-encoder to evaluate query-document pairs. The release describes using those modules together: embeddings find initial candidates, and a reranker can then refine their order. Qwen3-Embedding-4B itself supplies the single-text representation. Its vector is not a reranker score, a verified statement or a generated answer. If the final workflow answers questions from retrieved material, the selected source passages remain the evidence passed to that later stage.", "editorialOverlay": { "copy_fields": null, "copy_sha256": null, "copy_source": null, "html_sha256": "554f030acda1f2876d98b880cc14bde3793429009a823c1158e61fa7993e5b77", "module_sha256": "ee9dc1c50b1e6a8fb7fa062030fcb4646b9aaaeb08be513b723a5a92a7290f06", "raw_model_unchanged": true, "schema": "ailoot-native-model-editorial-v1" }, "locale": "en", "methods": [ { "html": "qwen/qwen3-embedding-4bQwen3-Embedding-4B represents text as semantic vectors for multilingual retrieval, code search, classification and clustering. This 4B model supports a 32K-token sequence length and custom vector dimensions from 32 to 2,560.
This description covers the model’s capabilities. Available in Ailoot inputs, outputs and options are listed for the selected method below.
Qwen3-Embedding-4B is the 4B text-embedding edition in Qwen’s embedding and reranking family. Built on the dense Qwen3 foundation model, it converts a text segment into a numerical representation of its meaning. Qwen identifies text retrieval, code retrieval, classification, clustering and bitext mining as application areas. It is useful when an application needs to find related passages, group similar records or compare text across languages. The model’s output is an embedding vector, not a conversational response or a rewritten document.
The exact edition has a 32K-token sequence length and an embedding dimension of up to 2,560. It supports custom output dimensions from 32 to 2,560 through Matryoshka Representation Learning. The chosen vector length affects how the representation is stored and compared, so keep it consistent between query and document vectors in one collection. Qwen3-Embedding-8B has a different model size and a 4,096-dimension maximum; neither that output size nor the 8B model’s identity transfers to this 4B edition. An existing index should retain its own consistent representation choices.
The producer describes support for more than 100 languages, including programming languages, with multilingual, cross-language and code retrieval. An illustrative help-centre search can compare a user’s question with passages written in several languages, then return the original passages that match the topic. A code search can compare a natural-language description with relevant functions and their documentation. Preserve meaningful headings, domain terms and identifiers when preparing text, and keep each represented passage connected to its source so the application can show the matching material.
Qwen3-Embedding-4B is instruction-aware. Qwen recommends giving each query a one-sentence description of the task, tailored to the intended retrieval relationship, and recommends English instructions for multilingual scenarios. The direct example places Instruct: before the task description and Query: before the query text. Its retrieval documents are encoded without that query instruction. A practical original instruction can specify finding documentation passages that explain a reported error or finding source-code excerpts that implement the described behaviour. This directs representation towards the application’s question rather than asking the model to write an answer.
The official examples encode query and document texts, compare their embeddings using cosine similarity and demonstrate pooling the final token’s hidden state followed by normalisation. In that lower-level example, the input is truncated at a chosen 8,192-token setting; the documented model sequence length remains 32K. For long material, decide which passages to represent and ensure that truncation does not silently discard the part the query needs. Store vectors with stable source references, compare the prepared query with the collection and return the associated passages in similarity order.
Qwen’s embedding model uses a dual-encoder approach, whereas its separate reranking models use a cross-encoder to evaluate query-document pairs. The release describes using those modules together: embeddings find initial candidates, and a reranker can then refine their order. Qwen3-Embedding-4B itself supplies the single-text representation. Its vector is not a reranker score, a verified statement or a generated answer. If the final workflow answers questions from retrieved material, the selected source passages remain the evidence passed to that later stage.
POST /v1/embeddingstext, token_idsembedding| Field | Type | Requirement | Values & limits | Default |
|---|---|---|---|---|
Fieldmodel | Typestring | RequirementRequired | Values & limits qwen/qwen3-embedding-4b | Default— |
Fieldinput | Typestring|string[]|integer[]|integer[][] | RequirementRequired | Values & limits Minimum items: 1 Conditions: Nonempty homogeneous batch. Token IDs are nonnegative integers. Input budget: Maximum: 32,768 · Text unit: UTF8 bytes · Token unit: token-array elements | Default— |
Fieldencoding_format | Typestring | RequirementOptional | Values & limits float | Defaultfloat |
Fielddimensions | Typeinteger | RequirementNot supported | Values & limits Minimum: 1 | Default— |
Fielduser | Typejson_value | RequirementOptional | Values & limits Conditions: Source allows this JSON metadata field; exact upstream semantics are model-specific. | Default— |
Fieldsession_id | Typejson_value | RequirementOptional | Values & limits Conditions: Source allows this JSON metadata field; exact upstream semantics are model-specific. | Default— |
Fieldtrace | Typejson_value | RequirementOptional | Values & limits Conditions: Source allows this JSON metadata field; exact upstream semantics are model-specific. | Default— |
Fieldprovider | Typeobject | RequirementOptional | Values & limits Conditions: provider.options rejected | Default— |
text_to_embedding
| Parameter | Contract |
|---|---|
Parametercontent_type | Contractapplication/json |
Parameterrequired | Contractmodel, input |
Parameterinput_shapes | Contractstring, string[], nonnegative_integer_token_array, nonnegative_integer_token_array[] |
Parameterencoding_format | Contractfloat |
Parameterdimensions_override | Contractfalse |
Parametermax_input_budget_units | Contract32768 |
Parametermultimodal_objects | Contractfalse |
Parametermaterial_types | Contracttext, token_ids |
Lifecycle: synchronous_json
Keep every vector index and dimension.
Multimodal model name does not admit image objects on this exact path.
Tariff references do not add request options. Supported parameters are listed for the selected method.
| Configuration | Rate schedule | Charge |
|---|---|---|
| ConfigurationBase token rates | Rate scheduleBase rate | Charge Input · Input tokens $0.00004 / 1K tokens Output · Output tokens $0 / 1K tokens Input · Cache read — / 1K tokens Input · Cache write — / 1K tokens Input · Cache write, 1 hour — / 1K tokens |
Ailoot reserves an estimated amount before a request. After the result, it settles the final charge from measured usage. Streaming settlement uses the final usage record. How reservation and settlement work →
set -eu\n: "${AILOOT_API_KEY:?Set your Ailoot API key}"\npython3 - <<'PY_REQUEST'\nimport json,os\nfrom pathlib import Path\np=json.loads('{"model": "qwen/qwen3-embedding-4b", "input": ["First document", "Second document"]}')\n\nPath("request.json").write_text(json.dumps(p,ensure_ascii=False))\nPY_REQUEST\ncurl --fail-with-body --silent --show-error https://api.ailoot.ai/v1/embeddings \\\n -H "Authorization: Bearer $AILOOT_API_KEY" -H 'Content-Type: application/json' \\\n --data-binary @request.json -D response-headers.txt -o result.json\npython3 - <<'PY_RESULT'\nimport json\nfrom pathlib import Path\nr=json.loads(Path("result.json").read_text())\nfor x in r.get("data",[]): print(json.dumps({"index":x.get("index"),"embedding":x.get("embedding")}))\nPY_RESULT\nfrom openai import OpenAI\nimport os\n\nclient=OpenAI(base_url="https://api.ailoot.ai/v1",api_key=os.environ["AILOOT_API_KEY"],max_retries=0)\nresult=client.embeddings.create(model='qwen/qwen3-embedding-4b',input=["First document","Second document"])\nprint(result)\nfor item in result.data:print(item.index,item.embedding)\njson
Keep every complete returned response, file and content part.
Keep every vector with its index. Vector count must equal input count; dimensions must be consistent.
No embedding job endpoint is introduced.
| Field | Type |
|---|---|
Fieldobject | Typestring |
Fieldmodel | Typestring |
Fielddata[].index | Typeinteger |
Fielddata[].embedding | Typefinite float[] |
Fieldusage | Typeobject |
| Response | Condition | Next step |
|---|---|---|
Response400 · invalid_request_error | ConditionInvalid required fields, material representation or parameter value | Next stepCorrect the indicated fields before a new creation request. |
Ailoot Ultra qwen/qwen3-embedding-4b
Use the method paths shown above.