Amazon Bedrock vs Amazon SageMaker — When to Use Each
A common AWS AI Practitioner (AIF-C01) exam trap: picking SageMaker when the scenario calls for Bedrock. This page walks through a representative practice question, then explains the decision rule that covers most exam variations.
Short answer
The correct answer is B. Amazon Bedrock.
Amazon Bedrock is a fully managed, serverless service that provides API access to foundation models from leading AI providers — including Anthropic, Meta, Cohere, AI21 Labs, Stability AI, Mistral, and Amazon's own Titan family — through a single, unified API. For a chatbot use case like this one, a team can call Bedrock's InvokeModel or Converse API with a prompt and receive generated output, with no infrastructure to provision, no instance sizing, and no model hosting to manage. Bedrock also offers managed add-ons directly relevant to chatbots: Knowledge Bases for retrieval-augmented generation (RAG) over enterprise data, Agents for multi-step tool use, and Guardrails for policy enforcement on inputs and outputs. Pricing is per-token (or optionally provisioned throughput), which matches the zero-infrastructure requirement in the scenario. The phrase 'multiple models from different providers through a single API' is the decisive clue — this is Bedrock's core value proposition. Any answer that requires you to deploy, host, or train a model is wrong for this scenario.
The Question
A company wants to deploy a customer-facing chatbot using a pre-trained foundation model. They need access to multiple models from different providers (Anthropic Claude, Meta Llama, Amazon Titan) through a single API and want to avoid managing any training or inference infrastructure. Which AWS service best fits this requirement?
Why B is correct
Amazon Bedrock is a fully managed, serverless service that provides API access to foundation models from leading AI providers — including Anthropic, Meta, Cohere, AI21 Labs, Stability AI, Mistral, and Amazon's own Titan family — through a single, unified API. For a chatbot use case like this one, a team can call Bedrock's InvokeModel or Converse API with a prompt and receive generated output, with no infrastructure to provision, no instance sizing, and no model hosting to manage. Bedrock also offers managed add-ons directly relevant to chatbots: Knowledge Bases for retrieval-augmented generation (RAG) over enterprise data, Agents for multi-step tool use, and Guardrails for policy enforcement on inputs and outputs. Pricing is per-token (or optionally provisioned throughput), which matches the zero-infrastructure requirement in the scenario. The phrase 'multiple models from different providers through a single API' is the decisive clue — this is Bedrock's core value proposition. Any answer that requires you to deploy, host, or train a model is wrong for this scenario.
Why the other options are wrong
SageMaker Training Jobs are for training or fine-tuning models from scratch. They require you to supply training scripts, choose instance types, and manage datasets — the opposite of "access pre-trained models with no infrastructure management."
SageMaker Endpoints host models you've built, brought, or deployed from JumpStart. They require you to configure instance types and manage deployment. They don't give you out-of-the-box, unified API access to third-party foundation models like Claude or Llama.
Amazon Comprehend is a managed NLP service for narrow tasks (sentiment analysis, entity recognition, key phrase extraction, language detection). It is not a generative AI service and does not expose foundation models for chat or content generation.
When to choose Bedrock vs SageMaker on AWS
Amazon Bedrock and Amazon SageMaker both sit inside the AWS AI stack, but they target different jobs-to-be-done. Confusing them is one of the most common mistakes on the AIF-C01 exam. Bedrock is purpose-built for CONSUMING foundation models as an API. You do not train models, you do not manage GPUs, and you do not pick instance sizes. You call a model like anthropic.claude-sonnet or meta.llama3-70b, pass a prompt, and receive a response. Bedrock's strengths are its multi-provider model catalog, managed RAG through Knowledge Bases, Agents for orchestration, Guardrails for safety policies, and serverless pay-per-token pricing. Common Bedrock use cases include chatbots, summarization, content generation, copilots, and RAG over private documents. SageMaker is a full end-to-end ML platform for the model lifecycle. You use it to label data (Ground Truth), experiment (Studio), train from scratch or fine-tune (Training Jobs), deploy (Endpoints), and monitor (Model Monitor). You bring your own data and models, or you start from SageMaker JumpStart templates. Common SageMaker use cases include training a custom image classifier, fine-tuning an open-source LLM on proprietary data, or deploying a traditional ML model (XGBoost, scikit-learn) at scale. The simplest heuristic: if you want to USE a foundation model, choose Bedrock. If you want to BUILD, TRAIN, FINE-TUNE, or HOST a model, choose SageMaker. There is a small overlap zone — SageMaker JumpStart offers pre-trained foundation models and fine-tuning workflows, and Bedrock supports customization and fine-tuning for selected models — but the heuristic holds on almost every exam question.
Ready to see how you'd score?
Take the free practice quiz and find out which AWS Certified AI Practitioner domains you need to focus on. No signup required.
Practice 5 similar questions
Same cert, same or adjacent domain. Use these after reviewing the explanation.