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.
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.
Frequently Asked Questions
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 $1 input and $5 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.
Top Alternatives
- Anthropic: Pick Anthropic's own API for the wider feature set; pick Bedrock if AWS must be the only data processor.