Amazon Bedrock and the Anthropic API are different ways to run Claude Code, operated and administered by different organizations. Bedrock routes model inference through AWS and uses AWS credentials and billing; the first-party Anthropic API routes requests through Anthropic and uses Anthropic credentials and billing. Neither is automatically more secure or cheaper: the right choice depends on your control requirements, model and region availability, and measured workload. A third option, Claude Platform on AWS, is accessed through AWS but is Anthropic-operated—not Bedrock.
What is different between Bedrock and the Anthropic API?
The main distinction is the operating boundary: who operates inference, where requests are routed, how you authenticate, and which account handles billing. Claude Code is the coding tool; these are alternative provider paths for its model requests, not identical services.
| Comparison | Amazon Bedrock | Anthropic API | Claude Platform on AWS |
|---|---|---|---|
| Inference operator | AWS | Anthropic | Anthropic |
| Authentication and administration | AWS credentials and IAM, with Bedrock account and model access setup | Anthropic Console/API credentials and organization controls | AWS IAM/SigV4 or service-specific API-key options, as described by AWS |
| Billing pathway | AWS account; Bedrock model pricing | Anthropic API usage billing | AWS-integrated billing for this separate service |
| Worth evaluating when | AWS governance, IAM, and cloud billing are central requirements | You want direct access to Anthropic-operated API services | You want the separate Anthropic-operated platform with AWS commercial integration |
| Key qualification | Model and feature availability depend on region and account | Uses Anthropic’s API surface, controls, and billing | Do not mistake AWS access or billing integration for AWS-operated inference |
AWS’s current comparison of Claude Platform on AWS and Bedrock describes differences in API surface, feature availability, authentication, and rate-limit ownership. Do not assume feature parity between the services. Anthropic’s Bedrock guide characterizes Bedrock as running on AWS-managed infrastructure with zero operator access by Anthropic personnel; that is Anthropic’s description of the service, not an independent audit conclusion.
How to set up Claude Code with Amazon Bedrock
Before you start
- An AWS account with Amazon Bedrock access.
- Access to the Claude model you intend to use, in the AWS account and region where you plan to run it.
- IAM permissions appropriate for Claude Code to invoke the model.
AWS requires first-time customers of Anthropic models to submit use-case details before their first invocation. AWS describes this as generally a one-time step at the account level, or through the AWS Organizations management account.
#1 Best Overall
Use the interactive setup
- Start Claude Code and choose the third-party platform option, then select Amazon Bedrock. Alternatively, launch Claude Code and run
/setup-bedrock. - Choose an authentication method: a detected AWS profile, Bedrock API key, access key and secret, or credentials already available in the environment.
- Select the AWS region and Claude model. Confirm that the model is enabled for your account in that region.
- Complete the wizard. Claude Code saves the selected settings in the user settings file.
Configure the Bedrock path manually
For a manual or managed deployment, set CLAUDE_CODE_USE_BEDROCK=1, provide credentials through the standard AWS credential provider chain, and configure a region if the active profile does not supply one. Use the current Claude Code Bedrock setup guide for the exact model identifiers supported by your account and region.
For a team rollout, pin model versions rather than relying on aliases such as opus or sonnet to keep pointing to a particular version. Defaults and model IDs can change, and Bedrock availability varies by account and region.
Use the Anthropic API instead
The first-party API path uses an Anthropic Console/API credential, Anthropic’s API endpoint, and Anthropic billing. Keep that credential in your organization’s approved secret-management system rather than in source code. An Anthropic API key is not a Bedrock credential: the two paths use different endpoints and authentication systems.
Rank #2
Security and data: what changes with each route?
Separate the service boundary from Claude Code safeguards
Bedrock means AWS operates model inference; the first-party API means Anthropic operates it. Claude Platform on AWS remains Anthropic-operated even though it is accessed through AWS. Decide which boundary fits your contracts and controls rather than treating every AWS-accessible option as the same service.
Recommended Free Tools
Claude Code provides permission modes, sandboxed shell execution, working-directory boundaries, and network approval controls. These controls can limit exposure, but they do not eliminate prompt-injection or approval risks. A command you approve may still have broader effects than expected, so review proposed code and commands before allowing them to run.
Understand the data-flow qualification
Anthropic’s commercial data policy—including for API and third-party platform customers—says it does not train generative models on code or prompts submitted under commercial terms unless the customer opts into data sharing for model improvement. This policy does not mean no information can ever be sent to Anthropic under every Claude Code configuration.
Anthropic’s Claude Code FAQ says that, when Bedrock or Vertex is configured and CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC is set, model API calls go directly to the provider endpoint and only essential telemetry is sent. The setting and provider configuration matter; do not interpret this as a blanket claim that all traffic in every setup goes only to AWS.
Use a threat-model checklist
- Operator: Which organization operates the inference service under the selected option?
- Routing: Where do prompts and code go, including telemetry and nonessential traffic?
- Identity: Can you enforce least-privilege IAM or equivalent organization controls?
- Data terms: What retention, training, and opt-in terms apply to your account and contract?
- Location: Do the available model and inference routes meet your regional or residency requirements?
- Operations: Can your team monitor and audit usage in the consoles and systems it already governs?
There is no established universal security winner between Bedrock and the Anthropic API. The relevant answer depends on your threat model, permissions, data settings, service terms, and required regional controls.
Free tools Windows power users keep installed
One-click scans. No signup required.
How much does Claude Code cost on Bedrock or the Anthropic API?
There is no single fixed Claude Code price for API deployments. Costs depend on token use, selected model, codebase size, and patterns such as automation, background work, and multiple sessions. Bedrock pricing is model- and tier-dependent; AWS publishes rates for input, output, caching, batch, and other usage modes where applicable. The region and inference route can also affect the price.
Rank #4
How to interpret Anthropic’s published API figures
Anthropic’s Claude Code cost guide, accessed in 2026, reports enterprise-deployment averages of around $13 per developer per active day and $150–$250 per developer per month, and says 90% of users are below $30 per active day. These are Anthropic-published usage figures, not a forecast for a particular team; Anthropic says costs vary widely and recommends piloting to establish a local baseline.
The dollar amount shown by Claude Code’s /usage command is a local estimate calculated from token counts at list prices or an organization-configured model pricing table. Check the relevant provider Console for authoritative billing.
Do not compare a subscription seat price to uncapped usage as if they were equivalent
Claude Code is included on paid Claude plans, but usage limits are shared with other Claude products. Anthropic’s public pricing page, accessed in 2026, lists Pro at $20 monthly or $17 per month with annual billing, Max from $100 per month, and Enterprise at $20 per seat per month with annual billing plus usage at API rates. Prices exclude applicable tax; plans and limits can change. A seat price includes plan limits, so it is not directly comparable to uncapped API or Bedrock consumption without accounting for the workload and included usage.
Build a fair provider estimate
- Run a pilot on representative repositories and developer tasks, including background tasks and automation.
- Record input, output, and cached token volumes where applicable, along with the models actually used.
- Compare the same model family where possible, then apply current provider rates for the intended region and usage tier.
- Include the cost of multiple sessions and other workload patterns your team expects, rather than extrapolating from one developer’s day.
- Validate estimates against the provider’s billing Console after the pilot.
Which route should your team choose?
Choose based on the operating boundary and the controls your organization needs, then validate model access and workload cost.
- Evaluate Bedrock if AWS IAM, account governance, and AWS billing align with your existing controls, and its model availability and regional routing meet your requirements.
- Evaluate the Anthropic API if direct access to Anthropic-operated API services and Anthropic’s organization and billing controls better fit your deployment.
- Evaluate Claude Platform on AWS separately if you want AWS commercial integration but an Anthropic-operated platform; do not treat it as Bedrock.
These are conditional choices based on documented service and control differences, not measured claims about relative latency, security, or savings. Check the current model IDs, region options, rates, and service terms before rollout; product availability, aliases, pricing, plans, and policies can change.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
