Model · Alibaba · Embeddings
Qwen3-Embedding (0.6B / 4B / 8B)
Alibaba's open text-embedding models in three sizes under Apache 2.0 — 0.6B is the headline for self-hosted search, with 4B and 8B for higher quality. 100+ languages including programming languages, 32K context, and adjustable vector size.
- Modality
- Embeddings
- License
- Apache 2.0 (Open)
- Parameter size
- 0.6B (headline); 4B and 8B siblings
- Context window
- 32,768 tokens 32K tokens on all three sizes.
- Released
- June 5, 2025
- Last verified
- October 28, 2026
- Runs locally
- Yes
What you need to run it
- Mac
- Yes. Any Apple Silicon Mac (M1 or newer) with at least 8 GB (16 GB recommended) of memory. Newer chips are faster, but memory is what decides whether it runs.
- Windows / Linux PC
- NVIDIA works best. A graphics card with at least 4 GB (12 GB recommended) of video memory.
- No graphics card
- Works. Runs on a normal computer's processor, just slower.
The 0.6B size runs on any recent laptop; the 4B and 8B sizes need a bigger Mac or a graphics card.
Numbers are for: 0.6B at 16-bit for the minimum; 4B for the recommended figures (4-bit MLX builds exist for 0.6B, 4B, 8B). Our estimate from the model’s size — the maker publishes no figure.
Strengths
- Apache 2.0 — no commercial restrictions
- The 8B ranked No. 1 on the MTEB multilingual leaderboard at launch (70.58, as of 2025-06-05)
- The 0.6B is small enough to run on a laptop or CPU server; it beats BGE-M3 on Alibaba's MTEB multilingual table (64.33 against 59.56)
- Adjustable vector size: 32 to 1024 dimensions on 0.6B, up to 2560 on 4B and 4096 on 8B
- 100+ languages plus code retrieval; instructions on the query side add about 1-5%
- Pairs with Qwen3-Reranker for a two-stage search pipeline
Weaknesses
- Text only — for images and documents use the newer Qwen3-VL-Embedding (January 2026)
- 4B and 8B need a graphics card for good throughput
- Queries need a one-sentence task instruction to get the best results
Try it
| Where | Type | Notes |
|---|---|---|
| Hugging Face (0.6B) | weights | Apache 2.0; 4B and 8B in the same collection |
| Sentence Transformers | local | pip install sentence-transformers |
| Text Embeddings Inference | local | Docker server for CPU or GPU |
| Alibaba Cloud Model Studio | hosted-api | API key required |
Official sources
- Model card (0.6B) model
- GitHub github
- Qwen blog announcement
- Paper paper
Change log
- — Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size and community builds (HF Qwen/Qwen3-Embedding-0.6B).
- — Initial entry. Added 2026-10-28 from the popularity sweep: 8.27M Hugging Face downloads/month for Qwen3-Embedding-0.6B. Released 2025-06-05 per the Qwen blog and the arXiv paper. Apache 2.0, 32K context, free to self-host. The multimodal Qwen3-VL-Embedding (Hugging Face repos from January 2026) is a sibling, not covered by this entry.
Esc