Amazon Bedrock

Managed access to foundation models from several providers on AWS

AI Models & PlatformsOverseasβ˜…β˜…β˜…β˜…β˜† 4.0

What is Amazon Bedrock?

Amazon Bedrock is AWS's managed service for running foundation models from several providers without operating any inference infrastructure. You call one API, and behind it sit models from Anthropic, Meta, Mistral, Cohere, Amazon and others, billed by usage and governed by the same identity and policy system as the rest of your account. Beyond raw inference it adds knowledge bases for managed retrieval, agents that call your functions, guardrails for filtering and evaluation jobs. It assumes an AWS account, the right region and a tolerance for console configuration.

Last updated: 2026-09-20. This site only provides an index; for exact features, pricing, and licensing, see the official website.

Key features

  • Managed API access to models from several providers
  • Serverless inference with no infrastructure to provision or patch
  • Knowledge bases for managed retrieval over your own documents
  • Agents that call your APIs and functions through tool use
  • Guardrails to filter topics, words and sensitive information
  • Model evaluation jobs and provisioned throughput for steady load

Pros & cons

Strengths

  • One API surface across several model vendors
  • Identity, billing and logging reuse your existing AWS setup
  • Retrieval and agent tooling arrive without extra services

Watch out for

  • An AWS account, IAM roles and region choices are prerequisites
  • Not every model is available in every region
  • Costs are hard to forecast without quotas and budgets set

Best for & use cases

managed model inference on aws, enterprise retrieval, agent building and multi-vendor model access

If you're comparing similar products, check the alternatives below, or browse all tools in the AI Models & Platforms category.

FAQ

What is Bedrock compared with SageMaker?

Bedrock is for using existing foundation models through a managed API. SageMaker is for building, training and hosting your own models. Many teams use Bedrock for inference and SageMaker only when they need custom training or their own container.

Which models are available?

Models from several vendors, including Anthropic, Meta, Mistral, Cohere and Amazon's own families, though availability differs by region and changes as new versions launch. Check the region list before designing around a specific model.

How is billing handled?

Usage is charged through your normal AWS bill, either per token or per provisioned throughput unit, and it counts toward existing budgets and cost tags. There is no separate vendor account to open, which is convenient but easy to overlook.