Amazon Bedrock is a fully managed service Generative AI service running in AWS that provides access to top-performing foundation models (FMs) from leading AI providers and Amazon via a single API. With Bedrock, you can rapidly prototype, securely customize models with your own data, and deploy generative AI applications without managing infrastructure.
NOTE: While the official AWS service is "Amazon Bedrock" I'll be using that term interchangeably with the most common "AWS Bedrock". As an AWS Generative AI Competency holder, Metal Toad is highly qualified to speak about AWS Bedrock and to help you if you'd like to get started with a professional engagement. =
AWS Bedrock's serverless design and integration with AWS Console and other services make it easy to build intelligent, enterprise-ready solutions while maintaining privacy, security, and responsible AI practices. If you are new to artificial intelligence you might check out our article, is it worth it to deploy Generative AI? An SMB perspective.
Bedrock has a number of significant capabilities, including:
This article will answer the following questions:
Key Amazon Bedrock Terms
- Foundation Model (FM) – A large, pre-trained AI model capable of generating diverse outputs like text, images, or embeddings.
- Base Model – A ready-to-use foundation model provided by Amazon or partners.
- Prompt – The input or instruction given to a model to generate a response.
- Model Inference – The process of generating an output from a prompt using a model.
- Embedding – A numerical representation of data used to measure similarity between inputs (e.g., text or image).
- Retrieval Augmented Generation (RAG) – Enhancing prompts with external data to improve model responses.
- Model Customization – Tailoring a model to specific tasks using fine-tuning or continued pre-training.
- Agent – An application that uses foundation models to interpret, reason, and act based on input.
- Provisioned Throughput – Reserved capacity for faster, more consistent inference performance.
- Guardrails – Safety mechanisms to filter or control model output and prevent inappropriate content.
Amazon Bedrock offers a flexible, pay-as-you-go pricing model designed to accommodate various generative AI workloads. Costs are primarily determined by how you use foundation models (FMs), whether through on-demand access, batch processing, or provisioned throughput.
The following table offers a quick summary of some prices, while a more complete list can be found here:
| Company | Model | Version | Per million input tokens | Per million output tokens |
| DeepSeek | DeepSeek | deepseek-chat | $0.27 | $1.10 |
| DeepSeek | DeepSeek | R1 | $0.55 | $2.19 |
| Amazon | Nova | Micro | $0.04 | $0.14 |
| Amazon | Nova | Lite | $0.06 | $0.24 |
| Amazon | Nova | Pro | $0.80 | $3.20 |
| Amazon | Nova | Premier | $2.50 | $12.50 |
| Anthropic | Claude | Claude 3.7 Sonnet | $3.00 | $15.00 |
| Anthropic | Claude | Claude 3.5 Sonnet | $3.00 | $15.00 |
| Anthropic | Claude | Claude 3.5 Haiku | $0.80 | $4.00 |
| Anthropic | Claude | Claude 3 Haiku | $0.25 | $1.25 |
| Anthropic | Claude | Claude 3 Opus | $15.00 | $75.00 |
| Meta | Llama | Llama 3.3 70B | $0.36 | $0.36 |
| Meta | Llama | Llama 3.2 90B | $0.72 | $0.72 |
| Meta | Llama | Llama 3.2 11B | $0.16 | $0.16 |
| Meta | Llama | Llama 3.2 3B | $0.15 | $0.15 |
| Meta | Llama | Llama 3.2 1B | $0.10 | $0.10 |
| Meta | Llama | Llama 3.1 405B | $1.20 | $1.20 |
| Meta | Llama | Llama 3.1 70B | $0.36 | $0.36 |
| Meta | Llama | Llama 3.1 8B | $0.11 | $0.11 |
| Meta | Llama | Llama 3 70B | $0.36 | $0.36 |
| Meta | Llama | Llama 3 8B | $0.30 | $0.60 |
| Meta | Llama | Llama 2 70B | $2.65 | $3.50 |
| AI21Labs | Jamba | Jamba 1.5 Large | $2.00 | $8.00 |
| AI21Labs | Jamba | Jamba 1.5 Mini | $0.20 | $0.40 |
| AI21Labs | Jamba | Jamba-Instruct | $0.50 | $0.70 |
| AI21Labs | Jurassic | Jurassic-2 Mid | $12.50 | $12.50 |
| AI21Labs | Jurassic | Jurassic-2 Ultra | $18.80 | $18.80 |
| Mistral | Mistral | Mistral Large | $4.00 | $12.00 |
| Mistral | Mistral | Mistral Small | $1.00 | $3.00 |
| Mistral | Mistral | Mixtral 8x7B | $0.45 | $0.70 |
| Mistral | Mistral | Mistral 7B | $0.15 | $0.20 |
Amazon Bedrock is designed to simplify access, customization, and deployment of powerful foundation models from leading AI providers (like Anthropic, Meta, Mistral, Cohere, and Amazon itself) without requiring you to manage any infrastructure.
Here’s how it works:
Unified API Access
Bedrock provides a single API to interact with multiple Foundational Models (text, image, and multimodal) letting you test and use different models without changing your application codebase.
Model Selection
You choose from a curated list of models based on your use case (e.g., text generation, summarization, image generation). Each model comes ready to use, and you can switch or compare them easily.
Prompt-Based Inference
You send a prompt (text, image, or structured input) to the model via the Bedrock API or console. The model processes the input and returns a generated response (inference).
Customization Options
You can personalize models using:
Agent & Workflow Automation
Bedrock supports building agents that can reason through tasks, call APIs, and use internal data systems to automate actions for your users.
Secure, Scalable Deployment
Using AWS’s serverless environment, you can deploy models into your applications with built-in support for scalability, enterprise-grade security, and compliance controls.
Performance & Cost Controls
Bedrock offers options like Provisioned Throughput to ensure consistent, high-speed inference and optimize for budget and latency.
Amazon Bedrock is readily available in the AWS console here: https://console.aws.amazon.com/bedrock/
Amazon Bedrock offers instant access to models from leading AI providers (e.g., Anthropic, Meta, Deepseek, and Amazon's own model: Nova) through a single API—no infrastructure management required. This accelerates prototyping and deployment. Being able to swap between models easily allows you pick the best model for your particular workload.
With some models you can customize models using your own data via fine-tuning, Retrieval-Augmented Generation (RAG), or embedding, and integrate them securely into applications with native AWS services like IAM, CloudWatch, and S3. We maintain information on which models accept fine tuning in our GenAI Model List spreadsheet.
With options like Provisioned Throughput, batch inference, and usage-based pricing, Bedrock helps you scale efficiently while controlling cost and performance—ideal for everything from Proof of Concepts to production workloads.
Amazon Bedrock offers a distinctive approach to generative AI by prioritizing model flexibility, enterprise readiness, and seamless AWS integration, but the space is rapidly evolving, with each provider pushing innovation forward.
NOTE: This back-and-forth and the relative stand alone nature of Generative AI workloads make them a perfect candidate to try out new cloud providers.
Key differentiators for Amazon Bedrock include:
Unlike many platforms that tie you to a single model (e.g., OpenAI in Azure or Google’s Gemini), Bedrock gives you a unified interface to top models from Anthropic, Meta, Mistral, Cohere, Stability AI, and Amazon—letting you choose the right model per task without vendor lock-in.
Bedrock is natively integrated into AWS. You get fine-grained access control (IAM), observability (CloudWatch), and scalability (Lambda, Step Functions) without managing infrastructure. Competing tools like ChatGPT or Claude either dumb access down or require more DevOps lift or separate billing and security domains.
With support for RAG, fine-tuning, and continued pretraining, Bedrock offers sophisticated customization options within a secure, managed environment—no GPU provisioning or container orchestration required.
Bedrock’s pricing model includes on-demand, batch, and provisioned throughput—offering predictable performance at scale. Many platforms charge flat rates or offer fewer optimization levers.
Amazon Bedrock makes it easy to build powerful generative AI applications using top foundation models without managing infrastructure. Whether you're prototyping or deploying at scale, here are five high-impact solutions you can create:
Automatically generate, format, and update documentation using AI models trained on your internal processes and terminology. From technical manuals to onboarding guides, Bedrock helps maintain consistent, up-to-date content with minimal human intervention.
With Retrieval-Augmented Generation (RAG), you can create tools that pull relevant information from internal data sources to deliver accurate, context-aware answers—perfect for internal wikis, legal teams, or compliance dashboards.
Produce marketing copy, product descriptions, reports, and more at scale. Customize models with your tone and branding to maintain consistency across thousands of outputs.
Extract key insights from long documents such as contracts, case studies, or medical records. Bedrock enables fast, accurate summarization and classification, helping teams focus only on what matters most.
Embed generative AI into your own apps—whether it’s for generating ideas, analyzing user input, or assisting with creative tasks. With Bedrock’s serverless architecture and multi-model access, you can tailor AI functionality to your specific use case.
NOTE: While Amazon Bedrock can be used to build chatbots and customer service agents, operating them at scale can become more expensive than hiring human agents—especially for high-volume, low-value queries where human performance still outpaces current AI capabilities. See our article how much does it cost to replace a human with genAI?
Want further proof? There is one exception to Proof of Concept funding offered by AWS for GenAI: no chatbots.
Amazon Bedrock is ideal for organizations that want to build generative AI applications quickly, securely, and at scale—without the overhead of managing infrastructure or committing to a single model provider. Here are a few archetypes that could benefit from Amazon Bedrock: