Qwen3-Reranker-8B

by Qwen

SOTA text cross-encoder reranker: instruction-aware, 100+ languages

1.8Mdl/month

8Bparams

HuggingFace Run it on your own data, free

Identifiers

Model ID

Qwen/Qwen3-Reranker-8B

Feature URI

mixpeek://reranker@v1/qwen3_reranker_8b_v1

Overview

Qwen3-Reranker-8B is the text reranking counterpart to Qwen3-Embedding, currently the top-scoring model on MTEB-R reranking benchmarks (69.76). It operates as a cross-encoder: concatenating query and document into a single input and running full bidirectional attention to produce a relevance score.

Like the embedding model, it supports instruction-tuned task conditioning: you can specify what makes a document relevant for your specific use case. On Mixpeek, it slots into the second stage of a two-stage retrieval pipeline: the embedding model retrieves candidates, and the reranker precision-sorts the top results.

Architecture

Cross-encoder architecture on Qwen3 8B backbone. Processes concatenated query-document pairs with full bidirectional attention. Outputs a scalar relevance score. Supports instruction prefixes for task-specific ranking criteria.

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: "text_embedding",
    version: "v1",
    parameters: { model_id: "Qwen/Qwen3-Embedding-8B" },
  },
});