Overview
The Qwen3 Embedding series represents a significant advancement over its predecessor, the GTE-Qwen series, in text embedding and reranking capabilities, built upon the Qwen3 foundation models.
Key Features
- Support for over 100 languages including various programming languages
- Robust multilingual, cross-lingual, and code retrieval capabilities
- Full spectrum of sizes from 0.6B to 8B for both embedding and reranking models
- Flexible vector definitions across all dimensions
- User-defined instructions to enhance performance for specific tasks, languages, or scenarios
- 8B size embedding model ranks #1 in MTEB multilingual leaderboard (as of June 5, 2025, score 70.58)
- Reranking model excels in various text retrieval scenarios
- Superior performance in multilingual and cross-lingual tasks
Model Variants
- Qwen3-Embedding-8B (text)
- Qwen3-VL-Embedding (multimodal - supports text, images, screenshots, and video)
- Multiple size options for different efficiency/effectiveness trade-offs
License
Open-sourced under Apache 2.0 license
Availability
Available on Hugging Face, ModelScope, and GitHub with published technical report and code.
