Qwen3-Reranker-0.6B-seq-cls

The Qwen3-Reranker-0.6B-seq-cls model is a text reranker from the Qwen3 Embedding series, converted to a sequence classification architecture for direct relevance scoring. It accepts a query-document pair and outputs a relevance score between 0 and 1, making it suitable for ranking retrieved passages in search and retrieval-augmented generation pipelines.

Key Strengths

• Multilingual support: Handles over 100 languages, including programming languages, with robust cross-lingual retrieval capabilities.
• Long context: Supports a sequence length of up to 32,768 tokens.
• Instruction-aware: Accepts a user-defined instruction to tailor relevance scoring for specific tasks, languages, or scenarios. Using instructions typically yields a 1–5% improvement over default behavior.
• Efficient size: 0.6B parameters, offering a balance of performance and computational cost.

Benchmark Performance

The model is part of the Qwen3 Embedding series, which achieved state-of-the-art results on multiple text embedding and ranking benchmarks. The 8B embedding model ranks No. 1 on the MTEB multilingual leaderboard (score 70.58 as of June 5, 2025). The reranking models in the series, including this 0.6B variant, excel in diverse text retrieval scenarios.