Alibaba · reranker
Qwen3 Reranker 8B
Qwen3 Reranker 8B is a downloadable relevance-scoring model for query-document pairs. It refines an initial search result set and does not create document vectors.
Downloadable checkpoint · apache-2.0
Official sources reviewed 2026-10-02. This page documents access and specifications; it does not report a model benchmark.
Task and deployment contract
Downloadable checkpoint; run the task-specific inference code linked below.
The 8B checkpoint has a documented 32K sequence window. This is a local model specification, not the per-document limit of the hosted qwen3-rerank API.
The model scores the relevance of a query and document using the documented instruction and yes/no scoring format. Preserve candidate IDs when turning scores into a sorted result list.
Score ranges are model-specific. Recalibrate relevance thresholds on labelled queries when replacing another reranker.
Self-hosting requires compute and an inference service. Downloadable weights do not make deployment, GPU time or operations free.
Billing and limits
No current provider rate for this exact configuration is recorded here. Open weights still require compute and operations; an unlisted rate does not mean a hosted API is free.
Recorded input window: 32,768 tokens. Apply the endpoint's stricter request and batch limits when applicable.
Fast Pivot — Access hundreds of AI models from a single API — competitive pricing, reliable uptime, enterprise-grade infrastructure.