Qwen3-Reranker-4B

Name: Qwen3-Reranker-4B
Author: Qwen

by Qwen

Mid-size 4B text reranker with strong multilingual accuracy

1.16Mdl/month

4Bparams

HuggingFace Run it on your own data, free

Identifiers

Model ID

Qwen/Qwen3-Reranker-4B

Feature URI

mixpeek://reranker@v1/qwen3_reranker_4b_v1

Overview

Qwen3-Reranker-4B is the mid-size option in the Qwen3 reranker family, offering stronger accuracy than the 0.6B variant while remaining significantly faster than the 8B model. It supports 100+ languages with instruction-following for custom relevance criteria, making it the sweet spot for production deployments that need both quality and throughput.

Architecture

Decoder-based cross-encoder built on the Qwen3 4B language model. Instruction-tuned for reranking tasks with support for user-defined relevance criteria via natural language prompts. Processes concatenated query-document pairs through causal attention layers with a relevance classification head.

Mixpeek SDK Integration

// Reranking is a retriever STAGE in Mixpeek, not an ingest-time extractor.
// The rerank stage runs a cross-encoder inference service; the shipped default
// is BAAI/bge-reranker-v2-m3. Pointing it at Qwen3-Reranker-4B means registering that
// model as a custom reranker plugin and naming it in feature_uri, which is an
// Enterprise path. Stage contract read from GET /v1/discovery/stages.
const retriever = await mx.retrievers.create({
  namespace_id: "my-namespace",
  retriever_name: "search-then-rerank",
  stages: [
    {
      stage_name: "candidates",
      stage_id: "feature_search",
      parameters: { limit: 100 },
    },
    {
      stage_name: "rerank_results",
      stage_id: "rerank",
      parameters: {
        inference_name: "BAAI__bge_reranker_v2_m3",
        query: "{{INPUT.query}}",
        document_field: "content",
        top_k: 10,
      },
    },
  ],
});