Skip to main content

Text rerank

A rerank model re-scores documents returned by initial retrieval, surfacing the most relevant results at the top.

Model overview

The gte-rerank model will be discontinued on May 30, 2026. Switch to qwen3-rerank. Official announcement.

Model

Max documents

Max input tokens per item

Max input tokens per request

Supported languages

Scenarios

qwen3.7-text-rerank

500

30,000

120,000 (recommended)

201 major languages and dialects, including Chinese, English, Spanish, French, Portuguese, Indonesian, Japanese, Korean, German, and Russian

  • Text semantic retrieval

  • Retrieval-augmented generation (RAG) applications

qwen3-vl-rerank

Text: 100

Image: 40

Video: 4

8,000

120,000

33 major languages, such as Chinese, English, Japanese, Korean, French, and German

  • Image clustering

  • Cross-modal search

  • Image retrieval

  • Video retrieval

qwen3-rerank

500

4,000

Over 100 major languages, such as Chinese, English, Spanish, French, Portuguese, Indonesian, Japanese, Korean, German, and Russian

  • Text semantic retrieval

  • Retrieval-augmented generation (RAG) applications

gte-rerank-v2

30,000

Over 50 languages, such as Chinese, English, Japanese, Korean, Thai, Spanish, French, Portuguese, German, Indonesian, and Arabic

  • Max input tokens per item: Maximum tokens per query or document. If a query or document exceeds this limit, the API returns an HTTP 400 error and does not truncate the input.
  • Max documents: Maximum documents per request. For qwen3-vl-rerank, the limit varies by document type (text, image, video, or mixed).
  • Max input tokens per request: Calculated as Query Tokens × Number of documents + Total document tokens. Must not exceed the per-request limit.

Input limitations

Model

Image

Video

qwen3-vl-rerank

JPEG, PNG, WEBP, BMP, TIFF, ICO, DIB, ICNS, and SGI (URL or Base64 supported)

MP4, AVI, and MOV (URL only)

Prerequisites

Obtain an API key and set the API key as an environment variable. To use the SDK: install the DashScope SDK.

HTTP

Replace {WorkspaceId} with your actual workspace ID.
This section describes the DashScope HTTP API.

Request

Request headers

Content-Type string (Required)The content type of the request. Must be application/json.Authorization string (Required)Authenticates the request with a Model Studio API key. Example: Bearer sk-xxxx.

Request body

model string (Required)The model name. Supported values: qwen3.7-text-rerank, qwen3-rerank, gte-rerank-v2, qwen3-vl-rerank.input object (Required)Input content.
Nest query and documents in the input object.

Properties

query string | object (Required)Query text (max 4,000 tokens).qwen3-vl-rerank supports two query formats:
  • String format: Pass a text string directly, e.g., "query": "What is a rerank model?".
  • Object format: Pass a dictionary specifying modality type and value as {"modality type": "input content"}. Supported types: text and image.
    • Text query: "query": {"text": "What is a text rerank model?"}
    • Image query: "query": {"image": "Image URL or Base64-encoded string"}
documents array (Required)Candidate documents to sort. Each element is a string.qwen3-vl-rerank accepts a dictionary or string per element: {"modality type": "text/image URL/video URL"}. Supported types: text, image, video.
  • Text: Key is text, value is a string. A plain string without a dictionary wrapper also works.
  • Image: Key is image, value is a URL or Base64 Data URI (data:image/{format};base64,{data}, where {format} is jpeg/png and {data} is the encoded string).
  • Video: Key is video, value must be a publicly accessible URL.
parameters object (Optional)Optional parameters.
Nest top_n and instruct in the parameters object.

Properties

top_n int (Optional)Number of top-ranked documents to return. Default: all documents. Values exceeding total documents return all.return_documents bool (Optional)Whether to include document text in results. Default: false. Supported models: gte-rerank-v2, qwen3-vl-rerank, qwen3-rerank.instruct string (Optional)Custom sorting instruction. Applies to qwen3.7-text-rerank, qwen3-rerank, and qwen3-vl-rerank. Guides the model to apply different sorting policies. Examples:
  • Q&A retrieval task (default): "Given a web search query, retrieve relevant passages that answer the query."
    • Focus: Find answers. Model prioritizes whether a document answers the query.
    • Example: Query "How to prevent a cold?" - Document "Washing hands frequently prevents colds" scores high; "A cold is a common illness" scores low (topically relevant but doesn't answer).
  • Semantic similarity sorting task: "Retrieve semantically similar text."
    • Focus: Determine semantic equivalence. Model evaluates whether query and document have consistent core meaning, regardless of wording or structure.
    • Example: In FAQ scenarios, "How do I change my password?" and "What if I forget my password?" are semantically similar (high score). Model focuses on whether both reflect the same user intent.
Write instructions in English. Default: Q&A retrieval task. More instructions are available in the model repository.fps float (Optional)qwen3-vl-rerank only. Controls video frame extraction count. Smaller values = fewer frames. Range: 0-1 (default: 1.0).
  • qwen3.7-text-rerank
  • qwen3-rerank
  • qwen3-vl-rerank
  • gte-rerank-v2
  • dashscope CLI
curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/rerank/text-rerank/text-rerank' \
--header "Authorization: Bearer $DASHSCOPE_API_KEY" \
--header 'Content-Type: application/json' \
--data '{
    "model": "qwen3.7-text-rerank",
    "input": {
        "query": "What is a rerank model?",
        "documents": [
            "Rerank models are widely used in search engines and recommendation systems. They sort candidate documents based on text relevance.",
            "Quantum computing is a cutting-edge field of computer science.",
            "The development of pre-trained language models has brought new advancements to rerank models."
        ]
    },
    "parameters": {
        "top_n": 2,
        "instruct": "Given a web search query, retrieve relevant passages that answer the query."
    }
}'

Response

request_id stringUnique request identifier for tracing and troubleshooting.output objectTask output.

Properties

results arraySorting results, ordered by relevance_score descending.

Properties

document dictOriginal document object. Returned only when return_documents is true. Format: {"text": "Original document text"}.index intDocument index in the input documents array.relevance_score doubleSemantic relevance between document and query. Range: 0.0-1.0 (higher = more relevant).
Scores are relative to the current request and cannot be compared across requests.
usage objectToken usage statistics.

Properties

prompt_tokens intReturned only by qwen3.7-text-rerank. The total number of tokens in the request input.total_tokens intThe total tokens consumed by the request, used for billing and audit.
code stringError code. Returned only for failed requests. See Error codes.message stringDetailed error message. Returned only for failed requests. See Error codes.
  • Successful response
  • Failed response
{
    "output": {
        "results": [
            {
                "index": 0,
                "relevance_score": 0.9334521178273196
            },
            {
                "index": 2,
                "relevance_score": 0.34100082626411193
            }
        ]
    },
    "usage": {
        "prompt_tokens": 79,
        "total_tokens": 79
    },
    "request_id": "85ba5752-1900-47d2-8896-23f99b13f6e1"
}

Use the SDK

Example

Call the rerank model API.
SDK parameter names match the HTTP API, but the structure differs. HTTP uses nested input and parameters objects; the SDK uses a flat structure. 
import dashscope
from http import HTTPStatus
# China (Beijing) region. Replace {WorkspaceId} with your actual workspace ID. The configuration varies by region.
dashscope.base_http_api_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"

def text_rerank():
    resp = dashscope.TextReRank.call(
        model="qwen3.7-text-rerank",
        query="What is a rerank model?",
        documents=[
            "Rerank models are widely used in search engines and recommendation systems. They sort candidate documents based on text relevance.",
            "Quantum computing is a cutting-edge field of computer science.",
            "The development of pre-trained language models has brought new advancements to rerank models."
        ],
        top_n=10,
        instruct="Given a web search query, retrieve relevant passages that answer the query."
    )
    if resp.status_code == HTTPStatus.OK:
        print(resp)
    else:
        print(resp)

if __name__ == '__main__':
    text_rerank()
Use qwen3-vl-rerank for multimodal reranking with an image query. 
import dashscope
from http import HTTPStatus
import json
# China (Beijing) region. Replace {WorkspaceId} with your actual workspace ID.
dashscope.base_http_api_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"

def vl_rerank():
    resp = dashscope.TextReRank.call(
        model="qwen3-vl-rerank",
        query={"image": "https://img.alicdn.com/imgextra/i3/O1CN01rdstgY1uiZWt8gqSL_!!6000000006071-0-tps-1970-356.jpg"},
        documents=[
            {"text": "Rerank models are widely used in search engines and recommendation systems. They sort candidate documents based on text relevance."},
            {"image": "https://img.alicdn.com/imgextra/i3/O1CN01rdstgY1uiZWt8gqSL_!!6000000006071-0-tps-1970-356.jpg"},
            {"video": "https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20250107/lbcemt/new+video.mp4"}
        ],
        top_n=2,
        return_documents=True
    )
    if resp.status_code == HTTPStatus.OK:
        print(json.dumps(resp, default=str, ensure_ascii=False, indent=4))
    else:
        print(resp)

if __name__ == '__main__':
    vl_rerank()

Sample output

The SDK wraps the HTTP response. For successful requests, code and message are always empty strings.
{
    "status_code": 200,
    "request_id": "4b0805c0-6b36-490d-8bc1-4365f4c89905",
    "code": "",
    "message": "",
    "output": {
        "results": [
            {
                "index": 0,
                "relevance_score": 0.9334521178273196
            },
            {
                "index": 2,
                "relevance_score": 0.34100082626411193
            }
        ]
    },
    "usage": {
        "prompt_tokens": 79,
        "total_tokens": 79
    }
}

Error codes

If the model call fails and returns an error message, see Error codes for resolution.

Rate limiting

Rate Limits.
API Usage
Text Generation
Image Generation
Video Generation
World models
3D Model Generation
Audio
  • Audio generation
Real-time Multimodal
Realtime API
Text Embedding
Translation models
Decision Model
More Models
Toolkit/Framework
TokenPlan
Model Production
File Management
Billing
Model Management
More