Qwen3 and DeepSeek-V3.1 models now available fully managed on AWS

Amazon Web Services announced that Qwen3 models from Alibaba and DeepSeek-V3.1 are now available globally as fully managed offerings in Amazon Bedrock. The addition expands the existing selection of open weight models in the service from providers including Meta, Mistral AI, and OpenAI. AWS said the move underscores its commitment to offering customers a broad selection of cutting-edge AI technologies so they can select the best models for their use case while benefiting from AWS security, privacy, and reliability. By using the models in Amazon Bedrock, customers benefit from enterprise-grade security including data encryption and strict access controls. Customers retain full control over their data, and AWS does not share model input and output data with model providers or use it to improve base models. Customers can also set up safeguards such as Amazon Bedrock Guardrails to detect and prevent hallucinations. The Qwen3 family includes four new open weight models capable of planning multistep workflows, integrating with tools and APIs, and handling long context windows. Qwen3-Coder models are optimized for complex software engineering scenarios. Qwen3-235B-A22B-Instruct-2507 uses a mixture of experts architecture for efficiency. Qwen3-32B is a dense model for limited computational resources. Customers can toggle between thinking and quick response modes. DeepSeek-V3.1 contains 685 billion parameters and is adept at software development, mathematical reasoning, and data analysis. According to Artificial Analysis, it demonstrates significant improvements over predecessors in reasoning and agentic skills. Luis Wang, director of Amazon Bedrock, said open weight models represent a vital frontier in AI innovation and that AWS has invested in making it the best place to run them securely, at scale, and cost effectively.