We use essential cookies and similar tools that are necessary to provide our site and services. We use performance cookies to collect anonymous statistics, so we can understand how customers use our site and make improvements. Essential cookies cannot be deactivated, but you can choose “Customize” or “Decline” to decline performance cookies.
If you agree, AWS and approved third parties will also use cookies to provide useful site features, remember your preferences, and display relevant content, including relevant advertising. To accept or decline all non-essential cookies, choose “Accept” or “Decline.” To make more detailed choices, choose “Customize.”
Customize cookie preferences
We use cookies and similar tools (collectively, "cookies") for the following purposes.
Essential
Essential cookies are necessary to provide our site and services and cannot be deactivated. They are usually set in response to your actions on the site, such as setting your privacy preferences, signing in, or filling in forms.
Performance
Performance cookies provide anonymous statistics about how customers navigate our site so we can improve site experience and performance. Approved third parties may perform analytics on our behalf, but they cannot use the data for their own purposes.
Allowed
Functional
Functional cookies help us provide useful site features, remember your preferences, and display relevant content. Approved third parties may set these cookies to provide certain site features. If you do not allow these cookies, then some or all of these services may not function properly.
Allowed
Advertising
Advertising cookies may be set through our site by us or our advertising partners and help us deliver relevant marketing content. If you do not allow these cookies, you will experience less relevant advertising.
Allowed
Blocking some types of cookies may impact your experience of our sites. You may review and change your choices at any time by selecting Cookie preferences in the footer of this site. We and selected third-parties use cookies or similar technologies as specified in the AWS Cookie Notice.
Your privacy choices
We display ads relevant to your interests on AWS sites and on other properties, including cross-context behavioral advertising. Cross-context behavioral advertising uses data from one site or app to advertise to you on a different company’s site or app.
To not allow AWS cross-context behavioral advertising based on cookies or similar technologies, select “Don't allow” and “Save privacy choices” below, or visit an AWS site with a legally-recognized decline signal enabled, such as the Global Privacy Control. If you delete your cookies or visit this site from a different browser or device, you will need to make your selection again. For more information about cookies and how we use them, please read our AWS Cookie Notice.
Chinese AI startup, DeepSeek has released cutting-edge family of large language models that has gained significant attention in the AI community for its impressive performance, cost-effectiveness, and open-source nature. DeepSeek offers a range of models including the powerful DeepSeek-V3 launched in December 2024. The model stands out for it’s innovative architecture, utilising techniques like Mixture-of-Experts (MoE) and Multi-Head Latent Attention (MLA) to achieve high performance with lower computational requirements causing a stir in the world of Large Language Models (LLMs) which has been dominated by giants like Anthropic and OpenAI. Subsequently released the reasoning-focused DeepSeek-R1, DeepSeek-R1-Zero with 671 billion parameters, and DeepSeek-R1-Distill models ranging from 1.5–70 billion parameters on January 20, 2025. These models stands out for their reasoning capabilities, achieved through innovative training techniques such as reinforcement learning (RL). They further added their vision-based Janus-Pro-7B model on January 27, 2025. The models are publicly available and are reportedly 90-95% more affordable and cost-effective than comparable models.
Image not found
DeepSeek Models
Why choose AWS for DeepSeek Deployment?
Enhanced Privacy and Security :- AWS offers robust privacy and security measures, enabling customers to securely store their data within their private AWS cloud when deploying open-source models.
Comprehensive Infrastructure Support :- AWS provides a comprehensive infrastructure platform, featuring a range of Amazon EC2 compute instances equipped with the latest CPUs, GPUs, and custom accelerators. Additionally, it offers industry-leading networking and storage solutions to cater to diverse performance and budget requirements.
Data Foundation :- AWS offers a suite of database services tailored to storing and searching vector embeddings. Notable examples include Amazon OpenSearch, Amazon Aurora PostgreSQL, and Amazon RDS PostgreSQL. As data becomes a pivotal differentiator in GenAI applications, AWS can your customers with purpose-built data stores that effectively enhance their use cases.
How to launch DeepSeek on AWS?
Deploy Deepseek-R1 Model directly in Bedrock console Go to the Amazon Bedrock console, choose Model access under Bedrock configurations in the left navigation pane. To access the fully managed DeepSeek-R1 model, request access for DeepSeek-R1 in DeepSeek. You’ll then be granted access to the model in Amazon Bedrock. Get started with the DeepSeek-R1 model in Amazon Bedrock .
Deploy DeepSeek-R1-Distill models in Amazon Bedrock and Amazon SageMaker AI. You can also deploy DeepSeek-R1-Distill models via Amazon Bedrock Custom Model Import. You can also use AWS Trainium and AWS Inferentia to deploy Distill models cost-effectively via Amazon Elastic Compute Cloud (Amazon EC2). a. Amazon Bedrock Marketplace for DeepSeek-R1 distilled Llama and Qwen models - How to deploy DeepSeek-R1-Distill using Amazon Bedrock Marketplace b. Amazon SageMaker JumpStart for DeepSeek-R1 distilled Llama and Qwen models - How to deploy DeepSeel-R1-Distill using Amazon SageMaker JumpStart c. Amazon Bedrock Custom Model Import for DeepSeek-R1 distilled Llama models - Refer d. Using Amazon EC2 Trn1 instances powered by AWS Trainium for the DeepSeek-R1 distilled Llama models – Refer e. DeepSeek models can be trained on Amazon SageMaker AI through Hugging Face integration - Refer
For publicly available models like DeepSeek-R1, you are charged only the infrastructure price based on inference instance hours you select for Amazon Bedrock Marketplace, Amazon SageMaker JumpStart, and Amazon EC2. For the Bedrock Custom Model Import, you are only charged for model inference, based on the number of copies of your custom model is active, billed in 5-minute windows.
Model :- DeepSeek-R1-Distill-Llama-8B This requires 2 Custom Model Units. Billing occurs in 5-minute windows, starting from the first successful invocation of each model copy. Price per minute will be $0.1570 for 2 Custom Model Units. 0.1570 * 5 minutes * 32 five minute windows = $25.12 + the model storage costs will be $3.90 for the month = $ 29.02/month
Price per 1,000 input tokens — $0.00135 Price per 1,000 output tokens — $0.0054
*above pricing are from Oregon Region
Securely deploy DeepSeek on AWS
Utilize IAM to Control Access: Since R1 is deployed through the marketplace, IAM can be employed to regulate access to Bedrock marketplace models. Control R1 access via the bedrock:CreateMarketplaceModelEndpoint and bedrock:Invoke Model functions.
Utilize a Trusted Model: The R1 model deployed through the marketplace utilizes the SafeTensors format. AWS packages the R1 model as a container and regularly scans the container for vulnerabilities.
Implement External Access Blocking: During model deployment, Bedrock enables the execution of models within an isolated VPC, thereby preventing any external access.
Refrain from accessing R1 directly: a. R1 is an unsafe model - The designers created an impressive model, but safety was not their primary concern. In contrast, Amazon Bedrock is built with security and privacy from the outset.
