Qwen3-Embedding-8B
8B open embedding model that topped the multilingual MTEB leaderboard
About
Qwen3-Embedding-8B is Alibaba's flagship open embedding model, built on the Qwen3 8B foundation with LoRA fine-tuning that keeps the base model's multilingual and reasoning abilities. It produces 4096-dimensional vectors (flexible via Matryoshka), handles 32K-token inputs, supports 100+ languages including code, and is instruction-aware for task-specific tuning. Alibaba reports an MMTEB score of 70.58, number one on the multilingual leaderboard at its June 5, 2025 release, under the Apache 2.0 license. The price of that quality is 8 billion parameters — you need real GPU infrastructure to serve it, and for English-only workloads smaller models get close for far less compute.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
MMTEB 70.58, #1 multilingual (Alibaba-reported, June 2025)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: June 5, 2025 release, 8B params, 4096 dims, 32K context, 100+ languages, Apache 2.0 all verified in Qwen's official blog and GitHub repo read this session.; MMTEB 70.58 #1 multilingual is Alibaba-reported — labeled as such.; GPU serving cost noted as the main practical limitation..