Qwen/Qwen3-Embedding-4B

Qwen3-Embedding-4B: A Specialized Text Embedding Model

Qwen3-Embedding-4B is a 4 billion parameter model from the Qwen family, purpose-built for text embedding and ranking. It is part of a series that includes various sizes (0.6B, 4B, 8B) for both embedding and reranking. This model leverages the strong multilingual capabilities, long-text understanding, and reasoning skills of the underlying Qwen3 foundational models.

Key Capabilities

• Text Embedding & Reranking: Designed for tasks like text retrieval, code retrieval, text classification, text clustering, and bitext mining.
• Multilingual Support: Supports over 100 languages, including various programming languages, offering robust multilingual and cross-lingual capabilities.
• Flexible Embedding Dimensions: Allows user-defined output dimensions ranging from 32 to 2560.
• Instruction-Aware: Supports user-defined instructions to optimize performance for specific tasks, languages, or scenarios, with a recommended 1-5% performance improvement when used.
• Long Context: Features a context length of 32,000 tokens.

Performance Highlights

While the 8B model in the series ranks #1 on the MTEB multilingual leaderboard (as of June 5, 2025, with a score of 70.58), Qwen3-Embedding-4B also demonstrates strong performance. On the MTEB multilingual benchmark, it achieves a mean score of 69.45, with notable results in instructional retrieval (11.56) and text retrieval (69.60). In English MTEB evaluations, it scores 74.60 mean, and on C-MTEB (Chinese), it achieves 72.27 mean.

Good for

• Developers needing a specialized model for generating high-quality text embeddings.
• Applications requiring robust multilingual and cross-lingual text understanding.
• Building systems for text retrieval, classification, clustering, and bitext mining.
• Scenarios where flexible embedding dimensions and instruction-aware processing are beneficial.