Qwen3-Vl-Embedding

Model IntroductionText

Qwen3-VL-Embedding is a multimodal embedding model from the Qwen series, designed to map both text and images into a shared vector space. It enables cross-modal retrieval and alignment without task-specific fine-tuning.

Key Features

  • Supports joint text-image embedding in a unified representation.
  • Strong zero-shot retrieval performance for image-to-text and text-to-image tasks.
  • Built on a vision-language backbone, preserving semantic alignment across modalities.
  • Produces dense vectors suitable for similarity search and clustering.
  • Compatible with standard vector databases and ANN indexes.

Use Cases

  • Cross-modal search engines (e.g., search images with text or vice versa).
  • Content recommendation based on visual and textual similarity.
  • Duplicate or near-duplicate detection across images and captions.
  • Zero-shot classification and clustering of multimodal data.
  • Building retrieval-augmented generation (RAG) systems that include images.

Benchmarks

Benchmark (Area)Score*
MMEB-V2 (Overall) (Overall)77.8 (Eval)
MMEB-V2 (Image Avg.) (Image)80.1 (Eval)
MMEB-V2 (Video Avg.) (Video)67.2 (Eval)
MMEB-V2 (VisDoc Avg.) (VisDoc)82.4 (Eval)
MMTEB (Text)67.9 (Eval)
SciFact nDCG@10 (Text Retrieval)0.7444 (Eval)
SciFact Recall@10 (Text Retrieval)0.882 (Eval)
SciFact MRR@10 (Text Retrieval)0.704 (Eval)

API request example

Endpoint /v1/chat/completions. Replace <API_KEY> with a key from the console.

Standard OpenAI Chat Completions. Swap `model` for the target slug.

curl https://api.ravenrouter.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <API_KEY>" \
  -d '{
  "model": "Qwen3-Vl-Embedding",
  "messages": [
    {
      "role": "user",
      "content": "Hello! Introduce yourself in one sentence."
    }
  ],
  "max_tokens": 64
}'

More models: API Reference → Examples by model

Qwen3-Vl-Embedding API — $0.29/1M input | RavenRouter AI