Amazon Web Services offers fully managed DeepSeek-R1 models on Amazon Bedrock

Amazon Web Services is the first cloud service provider to offer DeepSeek-R1 as a fully managed service

Further expands the ways customers can use DeepSeek-R1 and its distilled version on Amazon Bedrock

Beijing - March 11, 2025 Amazon Web Services announced that DeepSeek-R1 is now available as a fully managed serverless large language model (LLM) on Amazon Bedrock. Amazon Web Services is the first cloud service provider to launch this model as a fully managed service. DeepSeek-R1 is one of a series of models launched by artificial intelligence startup DeepSeek. It is a publicly available model with high-precision complex reasoning capabilities and deep understanding of context when handling complex tasks.

DeepSeek has been the focus of industry attention in the past few months. Its rapid development and its training technology have been widely reported by the global media. These technologies reportedly reduce the cost of its models by 90-95% compared with similar products, making them more cost-effective .

This launch further expands the ways customers can use DeepSeek-R1 and its distilled versions (i.e., small models trained similar to DeepSeek-R1) on Amazon Bedrock . AWS customers can now easily access fully managed, serverless DeepSeek-R1 models on Amazon Bedrock and apply them to enterprise-scale deployments.

Any customer can use its powerful features to solve complex problems, write code, process data, analyze data, etc. The fully managed approach allows customers to not worry about any complex technical settings or operations behind the scenes. By using DeepSeek-R1 on Amazon Bedrock , customers can also benefit from Amazon Web Services' enterprise-level security, including data encryption and strict access control, to ensure customer data privacy and compliance. Customers have full control over their data and can set up Amazon Web Services recommended protections such as Amazon Bedrock Guardrails to detect and prevent model hallucinations.

Previously, DeepSeek-R1 was launched on the Amazon Bedrock Marketplace , and customers can run the model using their own managed infrastructure. In addition, customers can also upload their own fine-tuned DeepSeek-R1 distilled version of the Llama model through the Amazon Bedrock custom model import feature and run it as a fully managed model. This feature allows customers to import and use their custom models and existing models through a single API.

Since the model became available in late January, thousands of customers have deployed the DeepSeek-R1 model using Amazon Bedrock’s custom model import capability .

If DeepSeek-R1 were personified, they would be an expert software engineer who could easily write code in Python, explain complex algorithms in plain language, or easily write papers on classic philosophers, or even translate project requirements into Chinese for international teams. It is particularly worth mentioning that they are available 24/7 and do not mind being asked hundreds of times.

“We are excited to bring DeepSeek-R1, a state-of-the-art model, to Amazon Bedrock, with significantly lower inference costs while delivering cutting-edge inference performance,” said Vasi Philomin, Vice President of Generative AI at Amazon Web Services. “Combined with features like Amazon Bedrock Guardrails, customers can implement AI safety guardrails while getting the built-in security and privacy protections provided by Amazon Bedrock. This fully managed implementation allows customers to use the model in a serverless, pay-per-token model, helping them scale from experimentation to production without having to manage any infrastructure.”

By making DeepSeek-R1 available as a fully managed serverless model in Amazon Bedrock, AWS continues to make the latest innovations in industry-leading generative AI models available to a broad range of customers, from startups to large enterprises, regardless of their technical capabilities.

By offering a wide selection of fully managed models from industry-leading AI companies, Amazon Web Services continues to enable companies to choose the right tools for their needs and become the easiest way for customers to build and scale generative AI.

Amazon Web Services strongly recommends that customers use Amazon Bedrock Guardrails and Amazon Bedrock model evaluation capabilities with DeepSeek-R1 to protect their generative AI applications. Amazon Bedrock Guardrails helps prevent applications from generating harmful or inappropriate content, just like guardrails on a highway prevent vehicles from straying off the road. This includes blocking offensive language, explicit content, or other content deemed inappropriate for end users, as well as identifying and removing personal data to protect user privacy.

In addition, customers can set specific rules based on company policies or industry regulations. Evaluation tools can help customers evaluate the performance of AI models under specific requirements. For more information, see Amazon Bedrock Guardrails and Amazon Bedrock Evaluation Tools .

To learn more about DeepSeek-R1’s capabilities, how to deploy models, and how to integrate DeepSeek-R1 with features like Amazon Bedrock Guardrails, visit the Amazon Web Services News blog or the relevant DeepSeek-R1 in Amazon Bedrock product page .

Customers can start using DeepSeek-R1 today in the Amazon Bedrock console .