Qwen3-Embedding-0.6B

Qwen3-Embedding-0.6B is the smallest text embedding model in the Qwen3 Embedding series, released June 2025 alongside the 4B and 8B sizes and a matching reranker line. With 28 layers, a 32K sequence length, and 1024 embedding dimensions, it supports over 100 languages including programming languages, with Matryoshka Representation Learning for custom output dimensions and instruction-aware encoding that Qwen reports improves most downstream tasks by 1 to 5 percent.

The 0.6B size is aimed at high-throughput retrieval workloads where embedding cost and latency dominate, such as large-scale RAG indexing and real-time semantic search. It ships under Apache 2.0 on Hugging Face as Qwen/Qwen3-Embedding-0.6B and works with sentence-transformers 2.7.0 and Transformers 4.51.0 or newer. It is distinct from the Qwen3-Embedding-8B flagship already in this catalog.