by Amazon Web Services

Claude in Amazon Bedrock review, pricing and verdict

Claude models run by AWS, billed and secured through your AWS account.

Best for
Cloud architects at AWS-first enterprises
Not for
Developers who need Anthropic's newest server-side tools or Files API on release day

By , Founder and editor

Should you use Claude in Amazon Bedrock?

Choose it when your company already runs on AWS and wants Claude under existing IAM roles, KMS keys and invoices. Skip it if you need structured outputs, the Files API or server-side tools on launch day, because the Messages endpoint omits them. Default quota starts at 2 million input tokens per minute.

AWS is the only operator and data processor of the inference stack, with Anthropic personnel locked out, which is the property Anthropic's own guidance points regulated buyers toward.

AWS distribution and existing enterprise contracts give it a strong channel, but the same Claude models are sold through Anthropic, Google Cloud and Microsoft Foundry, so the moat is procurement convenience rather than exclusive model access.

Also for

  • Security leads who need Claude under existing IAM and audit controls
  • Cloud and platform engineers at companies that already standardise on AWS
Show 3 more
  • Security and compliance leads in healthcare, finance and government who need IAM, KMS keys and audit trails around a frontier model
  • Backend engineers who must keep Claude traffic on a private network through AWS PrivateLink
  • AI product teams that want Guardrails or Knowledge Bases wrapped around Claude

Also not for

  • Solo builders who want a simple monthly plan instead of per-token billing

Main trade-off

The newer Messages endpoint omits several Anthropic features, so teams that depend on server-side tools or the Files API cannot run them here.

Claude in Amazon Bedrock pricing

Server plans

Claude in Amazon Bedrock server plans
Plan
Claude Haiku 5.5 (Global)Model API · $0.10 input / $0.50 output per 1M tokens. The smallest current Claude on Bedrock, for high-volume classification and routing. Prompts past the long-context threshold bill at $0.50 / $2.50. · Global endpoint
Claude Haiku 4.5 (Global)Model API · $1 input / $5 output per 1M tokens. Previous small Claude, ten times the price of Haiku 5.5 on AWS's list. · Global endpoint
Claude Sonnet 5.5 (Global)Model API · $2 input / $10 output per 1M tokens. Mid-tier model for coding and agent work at scale. · Global endpoint
Claude Sonnet 5 (Global)Model API · $2 input / $10 output per 1M tokens. Previous Sonnet generation, open to all Bedrock customers. · Global endpoint
Claude Opus 5.5 (Global)Model API · $4 input / $20 output per 1M tokens. Latest Opus for long-running coding and knowledge work; access set per model. · Global endpoint
Claude Opus 5 (Global)Model API · $5 input / $25 output per 1M tokens. Previous Opus generation; batch rates are half. · Global endpoint
Claude Opus 4.8 (Global)Model API · $5 input / $25 output per 1M tokens. Older Opus, open to all Bedrock customers. · Global endpoint
Claude Fable 5.1 (Global)Model API · $10 input / $50 output per 1M tokens. Frontier tier for ambitious coding, scientific and enterprise work; open to all Bedrock customers. · Global endpoint
Claude Fable 5 (Global)Model API · $10 input / $50 output per 1M tokens. Previous Fable generation, open to all Bedrock customers. · Global endpoint
Claude Haiku 4.5 (Global, Batch)Model API · $0.5 input / $2.5 output per 1M tokens. Half-price asynchronous batch inference. · Global endpoint
Claude Opus 5 (Global, Batch)Model API · $2.5 input / $12.5 output per 1M tokens. Half-price asynchronous batch inference. · Global endpoint
Claude Mythos 5.1 (Global)Model API · $10 input / $50 output per 1M tokens. Gated: invitation only, so it is not open to all Bedrock customers. · Global endpoint
Minimum commitment
None on-demand; Reserved tier uses 1-month or 3-month terms
Billing
Pay as you go · Committed spend
How the pricing works, in full

Per 1M tokens (input / output) on the global endpoint, from the AWS Price List file published 2026-10-08: Claude Haiku 5.5 $0.10 / $0.50, Claude Haiku 4.5 $1 / $5, Claude Sonnet 5.5 $2 / $10, Claude Opus 5.5 $4 / $20, Claude Opus 5 $5 / $25, Claude Fable 5.1 $10 / $50. Regional endpoints add 10% (Opus 5.5 lists at $4.40 / $22.00 in us-east-1). Haiku 5.5 prompts that cross the long-context threshold list at $0.50 / $2.50. Batch inference is 50% cheaper on select models, and AWS lists Standard, Flex, Priority and Reserved tiers. Reserved capacity bills per TPM-hour on 1-month or 3-month terms. The price_min figure is Haiku 5.5 input. Compare other options under inference hosting.

Costs that are easy to miss

  • Bedrock Guardrails bill separately per text unit; AWS's own worked example prices content filters and denied topics at $0.15 per 1,000 text units each.
  • Knowledge Bases charge for index storage on top of model tokens; AWS's worked example uses $5.00 per GB.
  • Choosing a regional endpoint for residency raises the token rate above the global price.

Claude in Amazon Bedrock regions and compliance

Uptime SLA
99.9%

Available regions

  • Africa (Cape Town)
  • Asia Pacific (Tokyo)
  • Asia Pacific (Seoul)
  • Asia Pacific (Osaka)
  • Asia Pacific (Mumbai)
  • Asia Pacific (Hyderabad)
  • Asia Pacific (Singapore)
  • Asia Pacific (Sydney)
  • Asia Pacific (Jakarta)
  • Asia Pacific (Melbourne)
  • Canada (Central)
  • Canada West (Calgary)
  • Europe (Frankfurt)
  • Europe (Zurich)
  • Europe (Stockholm)
  • Europe (Milan)
  • Europe (Spain)
  • Europe (Ireland)
  • Europe (London)
  • Europe (Paris)
  • Israel (Tel Aviv)
  • Middle East (UAE)
  • South America (Sao Paulo)
  • US East (N. Virginia)
  • US East (Ohio)
  • US West (N. California)
  • US West (Oregon)

Compliance certifications

  • ISO
  • SOC
  • CSA STAR Level 2
  • HIPAA eligible
  • GDPR
  • FedRAMP High (AWS GovCloud US-West)

As stated by the provider. Ask for the current report before you rely on one.

Managed services

  • Amazon Bedrock Guardrails
  • Amazon Bedrock Knowledge Bases
  • Amazon Bedrock AgentCore
  • Cross-region inference profiles
  • Batch inference
  • Reserved tier capacity
  • AWS PrivateLink
  • CloudWatch and CloudTrail logging

Access methods

  • Web console
  • API
  • CLI
  • Dashboard

Networking

  • Private networksSupported

What Claude in Amazon Bedrock does

Claude in Amazon Bedrock is the Amazon Web Services managed inference service for Anthropic's Claude models, offered through endpoints in 27 AWS regions. Anthropic's documentation states that it runs on AWS-operated infrastructure with zero operator access, and access, billing and audit logging all go through the customer's AWS account.

  • Global and regional endpoints

    The global endpoint routes dynamically across regions at the base rate, while regional endpoints pin inference to one AWS region for data residency, and US, EU, JP and AU inference profiles cover whole geographies.

  • Zero operator access

    Anthropic's documentation says the Messages endpoint runs on AWS-managed infrastructure that Anthropic personnel cannot access, so prompts stay inside the AWS security boundary.

  • Anthropic-shaped Messages API

    The endpoint follows bedrock-mantle.{region}.api.aws/anthropic/v1/messages with standard SSE streaming, an AnthropicBedrockMantle SDK client and three sign-in paths: a Bedrock service role, IAM assumed roles with 12-hour sessions, or bearer tokens.

  • High default quota

    The default limit is 2 million input tokens per minute, and you can request up to 5 million input and 500,000 output tokens per minute without extra Anthropic approval.

Show all 6 features

  • AWS-managed guardrails and agents

    Bedrock Guardrails, Knowledge Bases and AgentCore (managed runtime, memory, identity, gateway and observability for agents that run up to eight hours) wrap Claude in AWS-native controls.

  • Caching, thinking and tool use

    Prompt caching, extended thinking, tool use (including the bash, computer use, memory and text editor tools) and citations are supported on the Messages endpoint.

Strengths and limits

Strengths

  • Billing, IAM, KMS keys and CloudTrail all stay in one AWS account, so an AWS-standardised company can add Claude without onboarding a separate vendor.
  • Checked model by model against AWS's price file, the global endpoint charged what Hokai stores for the direct route on four current models, so choosing AWS adds no cost when residency is not a requirement.
  • New flagships arrive quickly: AWS published launch posts for two flagship Claude releases in September 2026, three weeks apart.
  • Anthropic's own comparison table lists AWS as the only inference data processor on Bedrock, a point regulated buyers weigh heavily.

Limits

  • The Messages endpoint does not support structured outputs, the Files API, server-side tools (code execution, web search, web fetch), the Message Batches API, Agent Skills or Claude Managed Agents, and it ignores the anthropic-beta header.
  • Feature timing follows the Bedrock release schedule, whereas Anthropic says Claude Platform on AWS typically gets features the same day as its own API.
  • Regional endpoints carry a premium, and the newest Fable model had regional endpoints in only one region in Anthropic's docs when checked, so strict residency can narrow the choice of model.
  • Two integrations with different APIs and model ID formats coexist (Converse and InvokeModel for Opus 4.6 and earlier, Messages for Opus 4.7 and later), which complicates migrations.

Claude in Amazon Bedrock alternatives

ServiceRegionsPricingComplianceChecked
Claude in Amazon Bedrock This pageAfrica · Asia Pacific · Canada · Europe +3Pay as you go · Committed spendHIPAA · GDPR · FedRAMP10 Oct
AnthropicPick Anthropic's own API for the wider feature set; pick Bedrock if AWS must be the only data processor.US · Europe · Asia PacificFree tier · Pay as you go · SubscriptionSOC 2 · ISO 27001 · ISO 42001 +19 Oct
Google Vertex AIGoogle Cloud's managed AI platform, now called Gemini Enterprise Agent Platform.US · Canada · South America · Africa +3Free tier · Pay as you go · Committed spendISO 27001 · SOC 2 · HIPAA26 Oct
LuminalCompiler-built inference for open models, as a serverless API or on-prem.--Pay as you go--23 Oct

Claude in Amazon Bedrock facts

Provider founded
2006
Headquarters
Seattle, WA, USA
Open source
No

What changed

  • Release Claude Opus 5.5 arrives on Amazon Bedrock source
  • Release Claude Fable 5.1 arrives on Amazon Bedrock source

All changes across Hokai this week

Questions about Claude in Amazon Bedrock

What does Claude in Amazon Bedrock actually cost?

You pay per token, with no seat fee. On the global endpoint, the small Haiku model is $0.10 input and $0.50 output per 1M tokens, the latest Opus is $4 and $20, and the top Fable tier is $10 and $50. A regional endpoint adds 10% to those rates, and cache reads and batch jobs are billed at lower rates on models that support them.

Can regulated teams use Claude in Amazon Bedrock?

AWS states that Bedrock is HIPAA eligible, in scope for ISO, SOC and CSA STAR Level 2, usable in line with GDPR, and FedRAMP High authorised in the AWS GovCloud (US-West) Region. Eligibility does not make a workload compliant on its own, since that depends on how you configure it. AWS also says data is encrypted in transit and at rest, can sit under customer-managed KMS keys, and is not used to improve base models. For the two newest Fable releases, AWS says customers eligible for Enterprise Frontier Safeguards can use zero data retention for internal use through 31 December 2026.

What are Claude in Amazon Bedrock's closest competitors?

Claude Platform on AWS is Anthropic's own service, run on Anthropic infrastructure, billed through AWS Marketplace and typically getting features the same day as the Claude API. The first-party route is Anthropic's own API, which Hokai lists as its own service. Anthropic's docs also list Claude on Google Cloud and on Microsoft Foundry for teams standardised on those clouds.

How does Claude in Amazon Bedrock compare to Anthropic's own API in 2026?

In AWS's price file, the global endpoint matched the direct-API list rate on the models checked, so the difference is features and control rather than price. The newer Bedrock endpoint drops several first-party features and does not support anthropic-beta headers. In exchange, AWS is the only data processor and the charge lands on your AWS invoice. That makes Bedrock the better fit when governance decides the purchase, and the direct API the better fit when new features matter most.

How do you set up Claude in Amazon Bedrock?

Turn on model access for the Claude models you want in the Bedrock console, since access criteria are set per model. Then install the SDK with pip install "anthropic[bedrock]", create an AnthropicBedrockMantle client with your AWS region and call messages.create with an ID such as anthropic.claude-opus-5-5. Anthropic's docs recommend a Bedrock service role for authentication, and the console playground lets you test a prompt before writing any code.

About Claude in Amazon Bedrock

Claude in Amazon Bedrock is how Amazon Web Services (AWS) serves models from Anthropic as a managed inference service. AWS operates the inference stack, and Anthropic's documentation states that Anthropic personnel have no access to it. You call it with AWS credentials, pay for it on your AWS bill and read its logs in CloudWatch and CloudTrail, which is the whole appeal for teams that already run their systems on AWS.

Read the full background

AWS's current price list covers Claude Fable 5.1, Claude Fable 5, Claude Opus 5.5, Claude Opus 5, Claude Sonnet 5.5, Claude Sonnet 5, the small Claude Haiku 5.5 and older releases such as Opus 4.8, Claude Opus 4.7 and Haiku 4.5. The Claude Mythos 5.1 and Claude Mythos 5 models are gated and need an invitation from Anthropic. Access rules are set per model, so a model that is open to every Bedrock customer today may not be for the newest release.

Two Claude integrations live inside Bedrock. The newer one serves the Messages API at a bedrock-mantle endpoint for Opus 4.7 and later, with the same request shape as Anthropic's first-party API. The older one uses the InvokeModel and Converse APIs with ARN-versioned model IDs for Opus 4.6 and earlier. Anthropic also runs a separate product, Claude Platform on AWS, where Anthropic operates the infrastructure and AWS Marketplace handles billing. It has its own capacity pool, so it is not the same thing as Bedrock.

Anthropic's own guidance is to choose Bedrock when AWS must be the sole data processor, or when a buyer needs FedRAMP High, IL4, IL5 or HIPAA-ready compliance. The price of that control is a narrower feature surface: the Messages endpoint leaves out structured outputs, the Files API, server-side tools and Agent Skills. Teams that need those go to the Anthropic API or to Claude directly, and developers can point Claude Code at Bedrock instead. To pick a model tier before you commit, read which Claude model to use, how to choose the right LLM and what Claude is for, or let Smart Match narrow the list for your workload.

Sources and guides

Official sources

Plans, regions, the uptime commitment and the compliance list on this page are read from Amazon Web Services's own pages and documentation. The verdict is Hokai's reading of them, not the provider's claim.

Hokai guides that mention Claude in Amazon Bedrock

Lists that include Claude in Amazon Bedrock

More AI Infrastructure Services on Hokai

  • PineconeManaged vector database for search, RAG and agent memory.
  • AgentMailEmail inbox API built so every AI agent can own, read and answer its own inbox
  • Amazon SESAWS email sending and receiving, metered per thousand messages
  • Cloudflare Email ServiceSend and receive email from Workers and agents, with unlimited inbound routing
  • PostmarkTransactional email API that keeps app mail apart from bulk sends
  • ResendEmail API for developers, with inboxes your AI agents can read and answer