Qwen3-Reranker-4B

The Qwen3-Reranker-4B is a text reranking model from the Qwen3 Embedding series, designed for ranking tasks such as retrieval, code retrieval, and classification. It is built on the dense foundational model of Qwen3, inheriting multilingual capabilities (supporting over 100 languages), long-context understanding (up to 32k tokens), and reasoning skills. The model is instruction-aware, allowing custom prompts to improve performance for specific tasks, languages, or scenarios. Using instructions typically yields a 1% to 5% improvement in downstream tasks.

Key Strengths

• Multilingual and cross-lingual support: Handles more than 100 languages, including programming languages, enabling robust multilingual and code retrieval.
• Long context window: 32k token sequence length for processing extensive documents and queries.
• Instruction-aware: Developers can provide custom instructions to tailor ranking behavior for specific use cases (e.g., classification, domain-specific retrieval).
• Scalable sizes: Part of a series with 0.6B, 4B, and 8B variants, allowing selection based on efficiency and effectiveness.

Benchmark Performance

The Qwen3 Embedding series achieves state-of-the-art results on text retrieval benchmarks. The 8B embedding model holds the No.1 position on the MTEB multilingual leaderboard as of June 5, 2025 (score 70.58). The reranking models, including the 4B variant, excel in various text retrieval scenarios, though specific reranking benchmark scores are not provided in the model card.

Model Details

Property	Value
Model Type	Text Reranking
Parameters	4B
Context Length	32k
Supported Languages	100+
Instruction Aware	Yes

For further details, refer to the official blog and GitHub repository.