Today, Amazon Bedrock and Amazon SageMaker AI have officially announced their support for the latest DeepSeek-R1 model launched by DeepSeek. This move will offer users more choices and flexibility to meet the diverse needs of AI model applications in various scenarios.

According to Amazon's official introduction, Amazon Bedrock is particularly suitable for teams that wish to quickly integrate pre-trained models through APIs. The platform provides convenient interfaces and a rich model library, enabling users to easily incorporate models like DeepSeek-R1 into their own applications without having to start from scratch. This significantly shortens the development cycle.

For teams looking to perform advanced customization, training, and deployment, Amazon SageMaker AI is an ideal choice. This platform offers powerful customization features, allowing users to fine-tune, train, and deploy models according to their specific needs. Additionally, SageMaker AI provides comprehensive infrastructure to ensure stable and efficient model operation.

Besides Amazon Bedrock and SageMaker AI, users can also deploy the DeepSeek-R1-Distill model cost-effectively through Amazon EC2 and the Amazon Trainium and Amazon Inferentia available on the SageMaker AI platform. This option gives users greater deployment flexibility, enabling them to select the most appropriate deployment plan based on their requirements and budget.

It is noteworthy that DeepSeek has introduced several significant models in recent months. In December 2024, DeepSeek launched the DeepSeek-V3 model, bringing new breakthroughs to the AI field. In January 2025, the company successively released DeepSeek-R1, DeepSeek-R1-Zero with 671 billion parameters, and the DeepSeek-R1-Distill model with parameter scales ranging from 1.5 billion to 70 billion. These launches further enriched DeepSeek's product line, offering users more options.

Furthermore, on January 27, 2025, DeepSeek introduced the Janus-Pro-7B model based on visual technology. This model excels in visual tasks, providing users with more accurate and efficient visual recognition capabilities. Compared to similar models, DeepSeek's models reduce costs by 90% to 95%, making them more cost-effective and competitive.