Qwen3-VL-Reranker-2B

by Qwen

Cross-modal reranker for text, image, and video retrieval

325Kdl/month

204likes

2.1Bparams

HuggingFace Run it on your own data, free

Identifiers

Model ID

Qwen/Qwen3-VL-Reranker-2B

Feature URI

mixpeek://reranker@v1/qwen3_vl_reranker_2b_v1

Overview

Qwen3-VL-Reranker-2B is a 2B-parameter cross-encoder reranker that scores (query, document) pairs where both the query and document can be text, images, screenshots, or video frames. Unlike embedding models that encode queries and documents independently, cross-encoders jointly attend to both, producing more accurate relevance scores at the cost of higher latency.

This model fills a critical gap in multimodal retrieval pipelines: first-stage retrieval uses fast bi-encoder embeddings to find candidates, then the reranker rescores the top-K candidates with cross-attention for higher precision.

Architecture

Cross-encoder architecture built on Qwen3-VL-2B-Instruct. Jointly processes query and document tokens with full cross-attention. Outputs a single relevance score per pair. Supports 30+ languages.

Mixpeek SDK Integration

import { Mixpeek } from "mixpeek";

const mx = new Mixpeek({ apiKey: "API_KEY" });

// Managed: create a collection over a bucket; Mixpeek runs this model's extractor
const collection = await mx.collections.create({
  namespace_id: "my-namespace",
  collection_name: "my-collection",
  source: { type: "bucket", bucket_ids: ["bkt_your_bucket"] },
  feature_extractor: {
    feature_extractor_name: "feature_search",
    version: "v1",
    parameters: { model_id: "Qwen/Qwen3-VL-Embedding-2B" },
  },
});