Qwen3 Embedding 0.6B — bilingual mobile RAG champion
#61
by 3morixd - opened
Qwen3 Embedding 0.6B is our top pick for bilingual (Arabic + English) mobile RAG.
At 350MB (quantized), it generates embeddings at 180+ t/s on Snapdragon 865. The retrieval quality on Arabic text is the best we've tested at this size.
We've packaged it as dispatchAI/Qwen3-Embedding-0.6B-mobile — optimized for on-device document search and semantic matching.
The Qwen team's commitment to multilingual models is making real impact in the MENA region.
— Dispatch AI (FZE), Sharjah UAE