Qwen3 Embedding 4B

Qwen3 Embedding 4B is a dense, 4-billion-parameter multilingual embedding model built on the Qwen3 foundation series. With 36 layers and a 32K-token context window, it excels at turning text and code into high-quality vector representations across more than 100 languages, including numerous programming languages. The model is engineered for instruction-conditioned embeddings, meaning developers can guide the representation process with natural language prompts to sharpen performance for specific tasks, languages, or domain scenarios. It outputs flexible vector dimensions ranging from 256 to 4096, enabling developers to balance quality against storage and speed requirements depending on the application.

The Qwen3 Embedding series extends the multilingual strengths and long-text reasoning capabilities of the Qwen3 base models into the embedding space. This lineage supports strong cross-lingual representation learning, making the model particularly effective for tasks like text retrieval, code retrieval, text classification, clustering, and bitext mining. The 8B sibling model achieved state-of-the-art ranking on the MTEB multilingual leaderboard as of mid-2025, which signals the series' competitive performance. Released under the Apache 2.0 license, the series offers a spectrum of model sizes from 0.6B to 8B, allowing developers to choose the right balance of efficiency and capability for production pipelines, research experiments, or enterprise-scale retrieval systems.