Claude on AWS
Claude on Amazon Bedrock: what AWS runs and how you get access
What Anthropic and AWS publish about calling Claude through Amazon Bedrock: access, endpoints, regions and what is not supported.
Claude on AWS Bedrock means using Anthropic's Claude models through Amazon Bedrock, where AWS runs the inference infrastructure and you reach Claude with your AWS account and AWS sign-in. Access is set model by model, and some Claude features are not available there. Bedrock has two endpoints, global and regional, and the regional one carries a 10 percent premium. Anthropic's other AWS option, Claude Platform on AWS, is operated by Anthropic instead.
plain answer
On Amazon Bedrock, AWS runs the servers that answer Claude
AWS lists Claude on its Amazon Bedrock pages, and Anthropic documents how to call it there. Anthropic's page for the Bedrock option says that Claude in Amazon Bedrock "runs on AWS-managed infrastructure". In plain terms, AWS runs the servers that answer your requests, and you reach them through your AWS account.
Anthropic documents a second way to reach Claude from AWS, called Claude Platform on AWS. Its page says: "Unlike Amazon Bedrock, where AWS operates the inference stack, Anthropic operates Claude Platform on AWS." The rest of this page is about the Bedrock option. The FAQ below sets the two side by side.
plain answer
Access starts with an AWS account and is set model by model
Anthropic lists one prerequisite: "An AWS account with Amazon Bedrock model access enabled for the Claude models you intend to use." It adds that access is decided for each Claude model separately, and that for a model its table does not mark as open, you check the Amazon Bedrock model access page in the AWS console. Its table marks the Mythos models as "limited availability" with the access note "Active (verification required)".
For sign-in, Anthropic describes three paths. The first, which it labels recommended, is a Bedrock service role: an AWS administrator creates the role and gives developers permission to pass it. The second is an assumed IAM role through your identity provider, with a 12-hour maximum session. The third is a bearer token, and Anthropic says "bearer tokens are the least preferred of the three authentication paths". Because the choice is a setup decision, it usually belongs with whoever administers your AWS account.
Anthropic publishes Bedrock packages for its software development kits, for example pip install -U "anthropic[bedrock]" for Python. Requests go to an address of the form https://bedrock-mantle.{region}.api.aws/anthropic/v1/messages. One limit to know before you plan tooling: Anthropic says its ant command line tool does not support Amazon Bedrock. For Claude Code, it says to use the latest version with Claude in Amazon Bedrock and to run claude update to upgrade.
plain answer
Bedrock has two Claude endpoints, and two models sit on only one
Anthropic documents Claude on Bedrock on two pages, one for each Bedrock endpoint. The page its address labels legacy covers the InvokeModel and Converse APIs, which are AWS's own calling styles, at the bedrock-runtime endpoint. The other covers the Messages API, the same request format Anthropic uses on its own API, at the bedrock-mantle endpoint.
The split matters for model choice. Anthropic says Claude Fable 5.2 and Claude Mythos 5.2 "aren’t available in Claude in Amazon Bedrock" and that Bedrock serves them on the bedrock-runtime endpoint only. If you need one of those two, plan around the older calling style.
For some newer models on that older endpoint, AWS needs a model ID with a region prefix such as us or eu, and a request with the plain model ID fails with an HTTP 400 error.
plain answer
Global endpoints add no premium and regional endpoints add 10 percent
Anthropic describes two endpoint types. Global means "dynamic routing across all available regions for maximum availability. No pricing premium." Regional means the endpoint resolves to the single AWS region you specify, "for data-residency requirements", and regional endpoints "carry a 10 percent pricing premium over global endpoints".
Anthropic's region table lists 27 AWS regions. Four are in the United States: us-east-1 (N. Virginia), us-east-2 (Ohio), us-west-1 (N. California) and us-west-2 (Oregon). The table also shows, region by region, which endpoint types are offered, so read your row before you promise a data location to anyone. Anthropic's older page also notes that model availability varies by region.
Anthropic gives a default quota of 2 million input tokens per minute, and says you can ask for up to 5 million input and 500,000 output tokens per minute without further Anthropic approval. It adds that AWS sets the requests-per-minute limits on the Bedrock side.
This page quotes no prices. AWS says on its pricing page: "Pricing is dependent on the modality, provider, and model. Please select the model provider to see detailed pricing."
plain answer
Some Claude features are not available on Bedrock
Anthropic lists what works on Claude in Amazon Bedrock: the Messages API, prompt caching, thinking, tool use, citations and structured outputs. Tool use includes the bash, browser use, computer use, memory and text editor tools.
It also lists what does not: URL sources for images and documents and the Files API; server-side tools such as code execution, web search and web fetch; Agent Skills, the MCP connector and programmatic tool calling; the Message Batches, Models, Admin, Compliance API and Usage and Cost API endpoints; Claude Managed Agents; and the server-side fallback parameter. The older bedrock-runtime page adds automatic prompt caching to its own not-supported list.
For anything not named here, Anthropic points readers to its features overview, which carries the full list with Amazon Bedrock availability. Check it against your design before you commit, because the lists change as Anthropic updates the documentation.
limits
Data handling, logging and support sit with AWS
Anthropic says: "Amazon Bedrock governs data handling for Claude in Amazon Bedrock." AWS's data protection page describes how. Each Region where Bedrock is available has one deployment account per model provider, owned and operated by the Bedrock service team, and AWS says: "Because the model providers don't have access to those accounts, they don't have access to Amazon Bedrock logs or to customer prompts and completions."
Anthropic says Claude in Amazon Bedrock "emits logs to both CloudWatch and CloudTrail" and recommends keeping activity logs on at least a 30-day rolling basis. On its older page it adds that turning on Bedrock's invocation logging "does not give AWS or Anthropic any access to your content."
For support, Anthropic says to contact AWS Support or your AWS account team and to include your AWS account ID and the request-id from any failed API response. This page does not cover compliance programs. For your own review, start from AWS's data protection page and the terms you hold with AWS and Anthropic.
Sources
Cited sources
- S1: Claude in Amazon Bedrock (Claude Platform Docs, bedrock-mantle) (no update date shown), read 2026-11-18
- S2: Claude on Amazon Bedrock (bedrock-runtime) (Claude Platform Docs) (no update date shown), read 2026-11-18
- S3: Claude Platform on AWS (Claude Platform Docs) (no update date shown), read 2026-11-18
- S4: Data protection (Amazon Bedrock User Guide, AWS) (no update date shown), read 2026-11-18
- S5: Amazon Bedrock pricing (AWS) (no update date shown), read 2026-11-18
- S6: Anthropic Claude on Amazon Bedrock (AWS product page) (no update date shown), read 2026-11-18
Straight answers
Frequently asked questions about Claude on Amazon Bedrock: what AWS runs and how you get access
Is Claude on Amazon Bedrock the same as the Claude API?
No, they are separate services that share a request shape. Anthropic says Claude in Amazon Bedrock uses "the same Messages API shape you use with the Claude API", but it runs on AWS-managed infrastructure, and several Claude API features, such as the Files API and Message Batches, are on Anthropic's not-supported list for Bedrock.
What is the difference between Claude in Amazon Bedrock and Claude Platform on AWS?
Who operates it is the main difference. Anthropic's page says AWS operates the inference stack on Bedrock and Anthropic operates Claude Platform on AWS, which "bills through AWS Marketplace". Anthropic's Bedrock page calls Claude Platform on AWS an Anthropic-operated alternative with AWS Marketplace billing and typically same-day feature access, and its Claude Platform page says it uses a separate capacity pool from both Anthropic's own API and Amazon Bedrock.
How much does Claude cost on Amazon Bedrock?
We do not quote prices because AWS sets and changes them. AWS's pricing page says pricing depends on the modality, provider and model, and that you select the model provider there for detail. Anthropic adds one rule on its side: regional endpoints carry a 10 percent premium over global endpoints. Anthropic also links its own cloud platform pricing section from its Bedrock pages.
Who can see my prompts when I use Claude on Bedrock?
AWS says model providers do not have access to Bedrock logs or to customer prompts and completions, and Anthropic says its personnel have no access to the inference infrastructure. Inside your own organization, AWS recommends setting up individual users with IAM Identity Center or IAM so that "each user is given only the permissions necessary to fulfill their job duties", and logging is a setting you control.
Which Claude models can I use on Bedrock?
It depends on the model and on your account. On the day we read Anthropic's page, it said Claude Fable 5.1, Fable 5, Opus 4.8, Sonnet 5, Opus 4.7 and Haiku 4.5 were open to all Amazon Bedrock customers, and that other models have their own criteria shown in the AWS console. Model lists change often, so check the console and Anthropic's table before you design around a model. Anthropic's table also lists Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5, and its Access column sends you to the AWS console for their access criteria.
Talk to the team that would do the work
Bring your requirements to a working session with the person who'll actually deliver.
Book a Discovery Call