Qwen3 LLMs Now Live on Amazon Bedrock & SageMaker

Amazon Web Services (AWS) announces the immediate availability of the Qwen3 family of large language models (LLMs) on Amazon Bedrock Marketplace and Amazon SageMaker JumpStart. This integration empowers developers to deploy Qwen3 models, ranging from 0.6B to 32B parameters, for generative AI applications. The move significantly expands AWS's foundation model offerings, particularly with the addition of Qwen3 on Amazon Bedrock.

Qwen3 represents the latest generation in the Qwen series, offering a comprehensive suite of dense and Mixture-of-Experts (MoE) architectures. These models deliver significant advancements in core LLM capabilities. They notably improve reasoning, instruction-following, and agent capabilities, setting a new benchmark for performance.

A unique feature is Qwen3's seamless switching between "thinking mode" and "non-thinking mode." This provides optimal performance across diverse scenarios. In thinking mode, Qwen3 significantly enhances mathematical problem-solving, code generation, and commonsense logical reasoning, surpassing previous models. Conversely, the non-thinking mode streamlines interactions for direct answers, boosting conversational flow and efficiency. This dual-mode functionality offers unparalleled flexibility for developers building dynamic AI applications.

Qwen3's Advanced Reasoning and Multilingual Capabilities

Qwen3 also excels in human preference alignment. This supports creative writing, nuanced role-playing, and complex multi-turn dialogues, delivering a more natural and engaging user experience. Furthermore, its robust multilingual support covers over 100 languages and dialects, enabling strong instruction following and translation capabilities globally. Its expertise in agent capabilities allows precise integration with external tools, achieving leading performance among open-source models in complex agent-based tasks. This makes Qwen3 a powerful choice for sophisticated AI agents.

Developers can now integrate Qwen3 into their existing AWS workflows with ease. Deployment is straightforward via the intuitive Amazon Bedrock Marketplace console or the comprehensive Amazon SageMaker JumpStart Studio UI. Programmatic deployment is also fully supported using the SageMaker Python SDK. This flexibility facilitates rapid experimentation and scalable production deployments for various generative AI use cases, from content generation to complex analysis.

The models support advanced functionalities like combining detailed reasoning traces with explicit tool calls within a single completion. This capability, accessible via the Invoke_Model API, streamlines complex agent-based tasks. It provides a more integrated and efficient approach to AI application development, eliminating the previous trade-off between chain-of-thought and deterministic tool use. AWS recommends specific GPU-based instance types for optimal performance, such as ml.g5-12xlarge for the 32B model.