SageMaker JumpStart adds Qwen3 embedding and reranking models
Amazon SageMaker JumpStart now offers Qwen3-VL-Embedding-2B and Qwen3-Reranker-4B, enhancing retrieval pipelines with multimodal understanding and refined result ranking. These models enable developers to build advanced search capabilities on AWS infrastructure, accepting diverse inputs for embedding and precise relevance scoring for reranking. Both models can be deployed with a few clicks via SageMaker Studio or the Python SDK.
- →Qwen3-VL-Embedding-2B model available for multimodal retrieval
- →Qwen3-Reranker-4B model for refining retrieval results
- →Simplified foundation model deployment in SageMaker JumpStart
Features (2) ›
- Qwen3-VL-Embedding-2B model available for multimodal retrieval
The Qwen3-VL-Embedding-2B model is now available in SageMaker JumpStart, capable of generating semantically rich vectors from text, images, screenshots, and videos. It supports over 30 languages and performs well on tasks like image-text retrieval and visual question answering.
- Qwen3-Reranker-4B model for refining retrieval results
The Qwen3-Reranker-4B model is added to SageMaker JumpStart to refine retrieval results by providing precise relevance scores for query and document pairs. It supports over 100 languages for text and code retrieval, classification, and clustering, with custom instruction support.
Enhancements (1) ›
- Simplified foundation model deployment in SageMaker JumpStart
Customers can now deploy these new Qwen3 models with minimal effort through SageMaker Studio or the SageMaker Python SDK. This integration aims to streamline the process of building specific AI use cases on AWS.
https://aws.amazon.com/about-aws/whats-new/2026/07/qwen3-search-retrieval-on-sagemaker-jumpstart/