SOTA Embedding Model for Agentic Workflows Now in Public Preview

Databricks launches Qwen3-Embedding-0.6B, a state-of-the-art multilingual embedding model for building retrieval-powered AI agents on enterprise data.

• •Built on the Qwen3 foundation with a 32k token context length and instruction-aware design that boosts retrieval performance by 1–5% via simple prompts
• •Outperforms most 0.6B-class models and surpasses flagship models from OpenAI and Cohere on MTEB multilingual and English v2 leaderboards
• •Supports Matryoshka Representation Learning (MRL), allowing embeddings to be truncated from 32 to 1024 dimensions at request time for cost/recall tradeoffs
• •First multilingual embedding model on Databricks, supporting 100+ languages and cross-lingual retrieval tasks
• •Runs on secure, fully managed serverless GPUs with autoscaling, available on Pay-Per-Token, AI Functions, and Provisioned Throughput surfaces