Introduction Understand Consider Choose Use Explore Resources

Choosing a generative AI service

Taking the first step

Purpose	Help determine which AWS generative AI services are the best fit for your organization.
Last updated	July 9, 2024
Covered services	Amazon Bedrock Amazon Bedrock Studio Amazon Q Business Amazon Q Developer Amazon SageMaker Amazon Titan foundation models Public foundation models

Introduction

Generative AI is a set of artificial intelligence (AI) systems and models designed to generate content—such as code, text, images, music, or other forms of data—that is often creative and original. These systems can produce new content based on patterns and knowledge learned from existing data. It is becoming increasingly used by organizations and businesses to:

Automate creative workflows — You can use generative AI services to automate the workflows of time-consuming creative processes such as writing, image or video creation, and graphic design.
Customize and personalize content — If you want to create audience-specific context, you can use generative AI services to generate targeted content, product recommendations, and customized offerings.
Augment data — You can use generative AI to synthesize large training datasets for other ML models to unlock scenarios where human-labeled data is scarce.
Reduce cost — Using synthesized data, content, and digital assets, you might be able to lower costs.
Faster experimentation — Generative models allow you to test and iterate on more content variations and creative concepts than would be possible manually.

This guide will help you select the AWS generative AI services and tools that are the best fit for your needs and your organization.

A twelve-minute video about building generative AI applications on AWS.

Understand

Amazon offers a range of generative AI services, applications, tools, and supporting infrastructure.

Which of these you use depends a lot of what you’re trying to do, how much choice you need in the foundation models you use, the degree of customization you’ll need in your generative AI applications, and the expertise within your organization.

Diagram showing the AWS generative AI stack, with infrastructure for foundational model (FM) training and inference at the bottom, tools to build large language models (LLMs) and other FMs above it, and finally applications that leverage LLMs and other FMs at the top of the stack.

Amazon Q — when you need pre-defined applications for your use case

At the top of Amazon's generative AI stack, Amazon Q generative AI-based applications make use of large language models (LLMs) and foundation models, but don’t require that you explicitly choose a model. Each of these applications is aimed at a different use case and all are powered by Amazon Bedrock.

Learn more about the primary Amazon Q generative AI–powered assistants currently available:

Amazon Q Business

Amazon Q Business is able to answer questions, provide summaries, generate content, and securely complete tasks based on the data in your enterprise systems. It supports the general use case of using generative AI to quickly start making the most of the information in your enterprise. Amazon Q Business lets you make English-language queries about that information and provides answers in a manner appropriate to your team’s needs. In addition, you can create lightweight, purpose-built Amazon Q Apps (currently in preview) within your Amazon Q Business Pro subscription.

Amazon Q Developer

Amazon Q Developer is aimed at helping you understand, build, extend, and operate AWS applications. The supported use cases include tasks that range from coding, testing, and upgrading applications, to diagnosing errors, performing security scanning and fixes, and optimizing AWS resources. The advanced, multistep planning and reasoning capabilities in Amazon Q Developer are aimed at reducing the work involved in common tasks (for example, performing Java version upgrades), and helping implement new features generated from developer requests.

Amazon Q Developer is also available as a feature in several other AWS services including AWS Chatbot, Amazon CodeCatalyst, Amazon EC2, AWS Glue, and VPC Reachability Analyzer.

Amazon Q in QuickSight

Amazon Q in QuickSight is aimed at meeting the needs of a specific use case—in this instance how to help you get actionable insights from your data by connecting Amazon Q to the Amazon Q QuickSight business intelligence (BI) service. You can use it to quickly build visualizations of your data, summarize insights, answer data questions, and build data stories using natural language.

Amazon Q in Connect

Amazon Q in Connect is designed to help you automatically detect customer issues, and provide your customer service agents with contextual customer information along with suggested responses and actions for faster resolution of issues. It combines the capabilities of the Amazon Connect cloud contact center service with Amazon Q. Amazon Q in Connect can use your real-time conversations with your customers, along with relevant company content, to automatically recommend what to say or what actions an agent should take to better assist customers.

Amazon Bedrock — when you need a choice of foundation models

If you're developing custom AI applications, need access to multiple foundation models, and want more control over the AI models and outputs, then Amazon Bedrock could be the service that meets your needs. Amazon Bedrock is a fully managed service, and it supports a choice of popular foundation models, including Anthropic Claude, Cohere Command & Embed, AI21 Labs Jurassic, Meta Llama, Mistral AI, Stable Diffusion XL and Amazon Titan.

Diagram showing Amazon Bedrock's broad choice of models, from Amazon and leading model providers.

In addition, Amazon Bedrock provides what you need to build generative AI applications with security, privacy, and responsible AI—regardless of the foundation model you choose. It also offers model-independent, single API access designed to offer the flexibility to use different foundation models and upgrade to the latest model versions, with minimal code changes.

Learn more about the key features of Amazon Bedrock:

Model customization: Model customization helps you deliver differentiated and personalized user experiences. To customize models for specific tasks, you can privately fine-tune FMs using your own labeled datasets. Custom models include capabilities such as fine-tuning, and continued pre-training using unlabeled datasets. The list of FMs for which Amazon Bedrock supports fine-tuning includes Cohere Command, Meta Llama 2, Amazon Titan Text Lite and Express, Amazon Titan Multimodal Embeddings, and Amazon Titan Image Generator. The list of supported FMs is updated on an on-going basis. In addition, Custom Model Import for Amazon Bedrock (currently in preview) allows you to bring your own custom models and use them within Amazon Bedrock.
Agents: Agents for Amazon Bedrock helps you plan and create multistep tasks using company systems and data sources—from answering customer questions about your product availability to taking their orders. You can create an agent by first selecting an FM and then providing it access to your enterprise systems, knowledge bases, and AWS Lambda functions to securely run your APIs. An agent analyzes the user request, and a Lambda function or your application can automatically call the necessary APIs and data sources to fulfill the request.
Guardrails: Guardrails for Amazon Bedrock evaluates user inputs and FM responses based on use case specific policies, and provides an additional layer of safeguards, regardless of the underlying FM. Using a short natural language description, Guardrails for Amazon Bedrock allows you to define a set of topics to avoid within the context of your application. Guardrails detects and blocks user inputs and FM responses that fall into the restricted topics.
Knowledge Bases: Knowledge Bases for Amazon Bedrock is a fully managed capability that helps you implement the entire Retrieval Augmented Generation (RAG) workflow from ingestion to retrieval and prompt augmentation without having to build custom integrations to data sources and manage data flows. Session context management is built in, so your application can support multi-turn conversations. You can use the Retrieve API to fetch relevant results for a user query from knowledge bases.
Converse API: Amazon Bedrock Converse API allows you to create conversational applications that send and receive messages to and from an Amazon Bedrock model. For example, you can create a chat bot that maintains a conversation over many turns and uses a persona or tone customization that is unique to your needs, such as a helpful technical support assistant.
Tool use (function calling): Tool use (function calling) gives a model access to tools that can help it generate responses for messages that you send to the model. For example, you might have a chat application that lets users find out the most popular song played on a radio station. To answer a request for the most popular song, a model needs a tool that can query and return the song information.
Amazon Bedrock Studio: Explore Amazon Bedrock Studio (in preview), a new SSO-enabled web interface that provides a way for developers across your organization to experiment with LLMs and other FMs, collaborate on projects, and iterate on generative AI applications. It offers a rapid prototyping environment and streamlines access to multiple foundation models (FMs) and developer tools in Amazon Bedrock. It also supports Knowledge Bases for Amazon Bedrock and Guardrails for Amazon Bedrock.

Amazon SageMaker — when you need to build custom models and have control over the full ML lifecycle, from data preparation to model deployment and monitoring.

Amazon SageMaker is designed to help you build, train, and deploy machine learning models, including FMs, at scale. Consider this option when you have use cases that can benefit from extensive training, fine-tuning and customization of foundation models. It will also help you through the sometimes-challenging task of evaluating which FM is the best fit for your use case.

Amazon SageMaker also provides infrastructure and purpose-built tools to help you throughout the ML lifecycle, including integrated development environments (IDEs), purpose-built distributed training infrastructure, governance tools, machine learning operations (MLOps) tools, inference options and recommendations, and model evaluation.

Explore key features of Amazon SageMaker that may help you determine when to use it:

Infrastructure for FM training and inference

AWS offers specialized, accelerated hardware for high performance ML training and inference.

Amazon EC2 P5 instances are equipped with NVIDIA H100 Tensor Core GPUs, which are well-suited for both training and inference tasks in machine learning.
Amazon EC2 G5 instances feature up to 8 NVIDIA A10G Tensor Core GPUs, and second generation AMD EPYC processors, for a wide range of graphics-intensive and machine learning use cases.
AWS Trainium is the second-generation ML accelerator that AWS has purpose-built for deep learning (DL) training of 100B+ parameter models.
AWS Inferentia2-based Amazon EC2 Inf2 instances are designed to deliver high performance at the lowest cost in Amazon EC2 for your DL and generative AI inference applications.

Consider

After you have decided on a particular generative AI service that is appropriate for your use case and expertise, you must choose the appropriate foundation model that gives you the best results for your particular use case.

Note that Amazon Bedrock has a model evaluation capability that can assist in evaluating, comparing, and selecting the best foundation models for your use case. Details on this capability are illustrated in this blog post.

Here are some critical factors to consider when choosing an appropriate foundation model for your use case:

Modality

Identify use cases/modality

What it is: Modality refers to the type of data the model processes—text, images (vision), or embeddings.

Why it matters: The choice of modality should align with the data that you're working with. For example, if your project involves processing natural language, a text-based model like Claude, Llama 2, or Titan Text G1, is suitable. If you want to create embeddings, then you might use a model like Titan Embeddings G1. Similarly, for image-related tasks, models such as Claude, Stability AI, and Titan Image Generator G1 are more appropriate. Your use case might also involve considering your data source and the support for data source connectors, such as those provided in Amazon Q Business.

Model size

Model Size

What it is: This criterion refers to the number of parameters in a model. A parameter is a configuration variable that is internal to the model and whose values can be estimated (trained) during the training phase from the given training data. Parameters are crucial as they directly define the model's capability to learn from data. Large models often have more than 50 billion parameters.

Why it matters: The number of parameters is a key indicator of the model's complexity. More parameters mean that the model can capture more intricate patterns and nuances in the data, which generally leads to better performance. However, these models are not only expensive to train, but also require more computational resources to operate.

Inference latency

What it is: Inference speed, or latency, is the time it takes for a model to process input (often measured in tokens) and return an output. This processing time is crucial when the model's responses are part of an interactive system, like an AWS Chatbot.

Why it matters: Quick response times are essential for real-time applications such as interactive chatbots or instant translation services. These applications depend on the model's ability to process and respond to prompts rapidly to maintain a smooth user experience. Although larger foundational models typically offer more detailed and accurate responses, their complex architectures can lead to slower inference speeds. This slower processing might frustrate users expecting immediate interaction.

To address this challenge, you can choose models optimized for quicker responses, even if it means compromising somewhat on the depth or accuracy of the responses.

Context window

Maximizing context window

What it is: In the context of large language models, a context window refers to the amount of text (in tokens) that the model can consider at any one time when generating responses.

Why it matters: Larger context windows enable the model to remember and process more information in a single run. This ability is particularly valuable in complex tasks such as understanding long documents, engaging in detailed conversations, or generating coherent and contextually accurate text over larger spans.

For example, in a conversation, a model with a larger context window would remember more of the earlier dialogue, allowing it to provide responses that are more relevant and connected to the entire conversation. This leads to a more natural and satisfying user experience, as the model can maintain the thread of discussion without losing context.

Pricing

Pricing considerations

What it is: The cost of using a foundational model, influenced by the model's complexity and the model provider’s pricing structure.

Why it matters: Deploying high-performance models often comes with high costs due to increased computational needs. While these models provide advanced capabilities, their operational expenses can be high, particularly for startups or smaller projects on tight budgets.

On the other hand, smaller, less resource-intensive models offer a more budget-friendly option without significantly compromising performance. It's essential to weigh the model’s cost against its benefits to ensure it fits within your project's financial constraints, helping you get the best value for your investment without overspending.

Fine-tuning

Fine-tuning and continuous pre-training capability

What it is: Fine-tuning is a specialized training process where a pre-trained model that has been trained on a large, generic dataset is further trained (or fine-tuned) on a smaller, specific dataset. This process adapts the model to particularities of the new data, improving its performance on related tasks. Continuous pre-training, on the other hand, involves extending the initial pre-training phase with additional training on new, emerging data that wasn't part of the original training set, helping the model stay relevant as data evolves. You can also use Retrieval Augmented Generation (RAG) to retrieve data from outside a foundation model and augment your prompts by adding the relevant retrieved data in context.

Why it matters: With fine-tuning, you can increase model accuracy by providing your own task-specific labeled training dataset and further specialize your FMs. With continued pre-training, you can train models using your own unlabeled data in a secure and managed environment. Continued pre-training helps models become more domain-specific by accumulating more robust knowledge and adaptability beyond their original training.

Data quality

Data quality is a critical factor in the success of a generative AI application. Here are some key factors to consider when it comes to the quality:

Relevance: Ensure that the data you use for training your generative AI model is relevant to your application. Irrelevant or noisy data can lead to poor model performance.
Accuracy: The data should be accurate and free from errors. Inaccurate data can mislead your model and result in incorrect outputs.
Consistency: Maintain consistency in your data. Inconsistencies in the data can confuse the model and hinder its ability to learn patterns.
Bias and fairness: Be aware of biases in your data, as they can lead to biased model outputs. Take steps to mitigate bias and help ensure fairness in your generative AI system.
Annotation and labeling: If your application requires labeled data, verify that the annotations or labels are of high quality and created by experts.
Data preprocessing: Prepare your data by cleaning and preprocessing it. This might involve text tokenization, image resizing, or other data-specific transformations to make it suitable for training.

Data quantity

Quantity along with quality goes hand in hand. It is important to remember:

Sufficient data: In most cases, more data is better. Larger datasets allow your model to learn a wider range of patterns and generalize better. However, the required amount of data can vary depending on the complexity of your application.
Data augmentation: If you have limitations on the quantity of available data, consider data augmentation techniques. These techniques involve generating additional training examples by applying transformations to existing data. For example, you can rotate, crop, or flip images or paraphrase text to create more training samples.
Balancing data: Ensure that your dataset is balanced, especially if your generative AI application is expected to produce outputs with equal representation across different categories or classes. Imbalanced datasets can lead to biased model outputs.
Transfer learning: For certain applications, you can leverage pre-trained models. Transfer learning allows you to use models that were trained on massive datasets and fine-tune them with your specific data, often requiring less data for fine-tuning.

It's also important to continuously monitor and update your dataset as your generative AI application evolves and as new data becomes available.

Quality of response

What it is: Lastly, the most essential criterion is the quality of response. This is where you evaluate the output of a model based on several quality metrics, including accuracy, relevance, toxicity, fairness, and robustness against adversarial attacks.

Accuracy measures how often the model's responses are correct (and you would typically measure this against a pre-configured standard or baseline).
Relevance assesses how appropriate the responses are to the context or question posed.
Toxicity checks for harmful biases or inappropriate content in the model's outputs.
Fairness evaluates whether the model's responses are unbiased across different groups.
Robustness indicates how well the model can handle intentionally misleading or malicious inputs designed to confuse it.

Why it matters: The reliability and safety of model outputs are paramount, especially in applications that interact directly with users or make automated decisions that can affect people's lives. High-quality responses ensure user trust and satisfaction, reducing the risk of miscommunication and enhancing the overall user experience, thus earning the trust of your customers.

Choose

Generative AI category	What is it optimized for?	Generative AI services
Amazon Q — when you need pre-defined applications for your use case	Making it easier for you to generate code and get answers to questions across business data, by connecting to enterprise data repositories to summarize the data logically, analyze trends, and engage in dialogue about the data.	Amazon Q Business Amazon Q Developer
Amazon Bedrock — when you need a choice of foundation models and capabilities	Allowing you to choose from the leading models, customize them with your own data, and build generative AI applications with the builder tools that Amazon Bedrock offers.	Amazon Bedrock Amazon Bedrock Studio
Amazon SageMaker — when you need to build, train, fine-tune and deploy foundation models at scale for your specific use case	Building, training, and deploying machine learning models, including foundation models, at scale.	Amazon SageMaker
Amazon FMs	These models are optimized to support a variety of multi-modal use cases such as text, image, and embeddings.	Amazon Titan
Infrastructure for FM training and inference	These services are optimized to maximize the price performance benefits in FM training and inference.	AWS Trainium AWS Inferentia

Use

Now that you have a clear understanding of the criteria you need to apply in choosing an AWS generative AI service, you can select which services are optimized for your needs and explore how you might get started using each of them.

Amazon Q

Get started with Amazon Q

Review your options for getting started with Amazon Q Business and Amazon Q Developer in either the AWS Management Console or the IDE.

Explore the guide
Work with Amazon Q

Use the Amazon Q Business and Amazon Q Developer User Guides, as well as the Amazon Q Business API Reference to learn how you can tailor Amazon Q to your business needs as well as help you understand, build, extend, and operate applications and workloads on AWS.

Explore the guides
Learn Amazon Q

Take this short, introductory AWS Skill Builder course to get a high-level overview of Amazon Q (requires registration).

Start the course

Amazon Q Business

What is Amazon Q Business?

Get an overview of Amazon Q Business, with explanations of what it is, how it works, and how to get started using it.

Explore the guide
Create a sample Amazon Q Business application

Learn how to create your first Amazon Q Business application in either the AWS Management Console or using the command line interface (CLI).

Explore the guide
Combine Amazon Q Business and AWS IAM Identity Center to build generative AI apps

Build private and secure enterprise generative AI apps with Amazon Q Business and AWS IAM Identity Center

Read the blog

Amazon Q Developer

What is Amazon Q Developer?

Get an overview of Amazon Q Developer, with explanations of what it is, how it works, and how to get started using it.

Explore the guide
Get started with Amazon Q Developer

Read this blog post to explore some key tasks that you can accomplish with Amazon Q Developer.

Read the blog
Working with Amazon Q Developer

Use the Amazon Q Developer Center for fast access to key Amazon Q Developer articles, blog posts, videos and tips.

Explore the Amazon Q Developer Center

Amazon Bedrock

What is Amazon Bedrock?

Learn how to use this fully managed service to make foundation models (FMs) from Amazon and third parties available for your use through a unified API.

Explore the guide
Frequently asked questions about Amazon Bedrock

Get answers to the most commonly-asked questions about Amazon Bedrock, including how to use agents, security considerations, details on Amazon Bedrock software development kits (SDKs), retrieval augmented generation, how to use model evaluation, and how billing works.

Read the FAQs
Guidance for generating product descriptions with Amazon Bedrock

Learn how to use Amazon Bedrock as part of a solution to automate your product review and approval process for an e-commerce marketplace or retail website.

Explore the solution

Amazon Bedrock Studio

What is Amazon Bedrock Studio?

Learn how you can use this web application to prototype apps that use Amazon Bedrock models and features, without having to set up and use a developer environment.

Explore the guide
Build generative AI applications with Amazon Bedrock Studio (preview)

This blog explains how you can build applications using a wide array of top performing models, evaluate, and share your generative AI apps within Amazon Bedrock Studio.

Read the blog
Building an app with Amazon Bedrock Studio

Use the Build mode in Amazon Bedrock Studio to create prototype apps that uses Amazon Bedrock models and features. You can also use the Build mode to try experiments not supported in the Explore mode playground, such as setting inference parameters.

Explore the guide

Amazon SageMaker

What is Amazon SageMaker?

Learn how you can use this fully managed machine learning (ML) service to build, train, and deploy ML models into a production-ready hosted environment.

Explore the guide
Get started with Amazon SageMaker

Learn how to join an Amazon SageMaker domain, giving you access to Amazon SageMaker Studio and RStudio on SageMaker.

Explore the guide
Get started with Amazon SageMaker JumpStart

Explore SageMaker JumpStart solution templates that set up infrastructure for common use cases, and executable example notebooks for machine learning with SageMaker.

Explore the guide

Amazon Titan

Amazon Titan in Amazon Bedrock overview

Get an overview of Amazon Titan foundation models (FMs) to support your use cases.

Explore the guide
Cost-effective document classification using the Amazon Titan Multimodal Embeddings Model

Learn how you can use this model to categorize and extract insights from high volumes of documents of different formats. This blog explores how you can use it to help determine the next set of actions to take, depending on the type of document.

Read the blog
Build generative AI applications with Amazon Titan Text Premier, Amazon Bedrock, and AWS CDK

Explore building and deploying two sample applications powered by Amazon Titan Text Premier in this blog post.

Read the blog

AWS Trainium

Overview of AWS Trainium

Learn about AWS Trainium, the second-generation machine learning (ML) accelerator that AWS purpose built for deep learning training of 100B+ parameter models. Each Amazon EC2 Trn1 instance deploys up to 16 AWS Trainium accelerators to deliver a high-performance, low-cost solution for deep learning (DL) training in the cloud.

Explore the guide
Recommended Trainium Instances

Explore how AWS Trainium instances are designed to provide high performance and cost efficiency for deep learning model inference workloads.

Explore the guide
Scaling distributed training with AWS Trainium and Amazon EKS

If you're deploying your deep learning (DL) workloads using Amazon Elastic Kubernetes Service (Amazon EKS), learn how you can benefit from the general availability of Amazon EC2 Trn1 instances powered by AWS Trainium—a purpose-built ML accelerator optimized to provide a high-performance, cost-effective, and massively scalable platform for training DL models in the cloud.

Read the blog

AWS Inferentia

Overview of AWS Inferentia

Understand how accelerators are designed by AWS to deliver high performance at the lowest cost for your deep learning (DL) inference applications.

Explore the guide
AWS Inferentia2 builds on AWS Inferentia1 by delivering 4x higher throughput and 10x lower latency

Understand what AWS Inferentia2 is optimized for and how it was designed from the ground up to deliver higher performance, while lowering the cost of LLMs and generative AI inference.

Read the blog
Machine learning inference using AWS Inferentia

Learn how to create an Amazon EKS cluster with nodes running Amazon EC2 Inf1 instances and optionally deploy a sample application. Amazon EC2 Inf1 instances are powered by AWS Inferentia chips, which are custom built by AWS to provide high performance and lowest cost inference in the cloud.

Explore the guide

Explore

Architecture diagrams

These reference architecture diagrams show examples of AWS AI and ML services in use.

Explore architecture diagrams
Whitepapers

Explore whitepapers to help you get started and learn best practices in choosing and using AI and ML services.

Explore whitepapers
AWS solutions

Explore vetted solutions and architectural guidance for common use cases for AI and ML services.

Explore solutions

Resources

Public foundation models

Supported foundation models are updated on a regular basis, and currently include:

Use Amazon Bedrock and Amazon SageMaker to experiment with a variety of foundation models, and privately customize them with your data. To explore generative AI quickly, you also have the option of using PartyRock, an Amazon Bedrock Playground. It is a generative AI app building playground that allows you to experiment hands-on with prompt engineering.

Warning Javascript is disabled or is unavailable in your browser.

To use the Amazon Web Services Documentation, Javascript must be enabled. Please refer to your browser's Help pages for instructions.

Document Conventions

Document history