Amazon Web Services (AWS) announced on the 3rd that it is providing the DeepSeek-R1 model on Amazon Bedrock and Amazon SageMaker AI.
DeepSeek launched the DeepSeek-R1, R1-Zero, and R1-Distill models on January 20, 2025, following the DeepSeek-V3 in December 2024. On January 27, it added the vision-based model Janus-Pro 7B. The DeepSeek models offer an expense efficiency of 90-95% compared to previous versions and possess excellent reasoning abilities through reinforcement learning techniques.
AWS integrates the pre-trained DeepSeek-R1 model as an API on Amazon Bedrock so that customers can utilize it and supports custom training and deployment through Amazon SageMaker. In particular, using AWS Trainium and AWS Inferentia allows for the economical operation of the DeepSeek-R1-Distill model.
Model deployment is possible through ▲ Amazon Bedrock Marketplace ▲ SageMaker Jumpstart ▲ Bedrock Custom Model Import ▲ AWS EC2 Trn1 instances. To ensure security and scalability, AWS creates a safe AI environment using Amazon Bedrock Guardrails.
AWS plans to actively support customers in utilizing AI technology safely and efficiently through the DeepSeek-R1 model.