Claude Code Bedrock; Bedrock Claude Code pricing; AWS Bedrock pricing for Claude; Bedrock Claude cost

Running Claude Code on Amazon Bedrock puts model usage inside your AWS billing environment. That can simplify purchasing and access management, but understanding the bill still requires more than checking a token price.

Your Bedrock Claude cost depends on the model, inference configuration, caching, and how many requests a coding agent makes to finish its work.

This guide explains Bedrock Claude Code pricing, where charges appear, and how to track spending across developers, teams, and projects.

How Does Claude Code Work on Bedrock?

Claude Code is the development tool; Amazon Bedrock provides access to the Claude model handling its requests.

A typical rollout requires model access, approved AWS authentication, IAM permissions, and a supported region. Pin model versions and test reporting before expanding access.

Anthropic’s official Bedrock setup guide covers configuration and inference-profile mappings. Record the endpoint and API you use, because attribution capabilities differ between deployment paths.

What Determines AWS Bedrock Pricing for Claude?

For token-based inference, costs follow the billable usage generated by your workload.

Consult the Amazon Bedrock pricing page for your exact configuration. Normalize rates before comparing them: prices per 1,000 tokens and per million tokens use different units.

Reference Model Rates

AWS’s published cost-optimization guide uses these reference figures:

These examples explain the calculation. They are not a complete catalog or a quote for every region, profile, or tier. Confirm the applicable current rates before budgeting.

A Simple Bedrock Claude Code Cost Calculation

Using the Sonnet reference rates, consider 100,000 uncached input tokens and 20,000 output tokens:

Five calls with identical usage would cost $3.00 in estimated token charges. Actual agent calls vary as context grows.

For cached workloads, calculate uncached input, cache writes, cache reads, and output separately. Do not charge cached tokens again as uncached input.

Keep the pricing version alongside usage records. Treat calculations as estimates until reconciled with AWS billing, including applicable discounts and additional charges.

Where Does Claude Code Bedrock Spending Appear?

Bedrock-backed model usage is billed through AWS, but filtering only for “Amazon Bedrock” can miss relevant charges.

For example, AWS documents Claude Sonnet 4.6 as billed through AWS Marketplace, with charges appearing under the model provider in the AWS bill and Cost Explorer. Check the model’s AWS documentation and your own billing records.

Separate three figures:

• Estimated inference cost: Calculated from tokens and rates.
• AWS-reported billed cost: The financial reconciliation reference.
• Total operating cost: Model charges plus supporting services.

Logging, storage, query services, gateways, and development compute can add costs when deployed. Track separately purchased Claude subscriptions independently; do not assume their allowances cover Bedrock inference.

How to Track Costs by Developer, Team, and Model

1. Establish Reliable Developer Identity

AWS’s IAM principal attribution captures the calling IAM user or role for billing analysis.

A shared role alone does not distinguish developers. Configure trusted session tags or suitable identity mappings, activate relevant cost allocation tags, and include caller identity in your CUR 2.0 export.

If a gateway uses one shared AWS session, retain authenticated developer ownership in gateway records or configure appropriate per-user role sessions.

2. Allocate Team and Project Costs

Application inference profiles support model-specific cost allocation for InvokeModel and Converse workloads on the Bedrock runtime endpoint.

Create profiles, attach team or project tags, activate them for billing, and ensure requests use the intended ARNs.

For Claude Code’s Invoke deployment, map selected model versions to those profiles. Check auxiliary models too. Profiles are model-specific, and other API paths require compatible attribution mechanisms.

3. Capture Session-Level Activity

Claude Code’s OpenTelemetry monitoring exposes token and approximate cost metrics with model and session attributes.

Use this telemetry to identify expensive workflows. With Bedrock authentication, attach accountable identity through managed configuration rather than treating an anonymous installation identifier as a verified employee.

For a broader approach, read how to track Claude and GPT costs by developer, team, and model.

4. Reconcile With AWS Billing

Compare telemetry and billing over equivalent periods, models, and usage categories. Investigate missing accounts, duplicate records, pricing differences, and reporting delays.

AWS’s request-metadata documentation distinguishes request logs from aggregated billing. Request metadata is not automatically a billing allocation tag, and billing lines do not provide a one-to-one request ledger.

Which Tracking Option Should You Use?

Combine billing tools with telemetry. One explains the charges; the other helps explain the work behind them.

Set Budgets and Reduce Waste

AWS Budgets supports cost alerts and configured actions. However, these depend on billing updates. AWS’s budget best practices describe at-least-daily billing refreshes.

A notification therefore is not a precise real-time cap. For daily developer limits, use tested request-level enforcement that handles concurrent requests, costs in flight, and approved exceptions. See our guide to Claude Code budgets by user and project.

To reduce spending:

• Evaluate cheaper models for routine work.
• Remove unnecessary files and oversized tool results from context.
• Measure caching and verify supported models and cache lifetimes.
• Limit retries and long-running agent loops.
• Compare service-tier cost and latency.

Use AWS’s prompt-caching guide, alongside our articles on smart model routing and runaway agent costs.

Where Opviera Fits

Opviera’s Bedrock cost-control guide describes routing supported Claude Code workflows through a gateway with developer attribution, project budgets, and model policies.

Pilot the integration, check endpoint and model coverage, include platform fees, and account for direct-access routes. Continue reconciling recorded estimates with AWS charges.

Frequently Asked Questions

Is Claude Code on Bedrock free?

Model inference is usage-based. Credits or negotiated arrangements apply according to their terms. Supporting services can create separate charges.

Is Bedrock cheaper than the direct Anthropic API?

Compare the same model, token categories, inference configuration, and commercial terms. Include operating costs rather than assuming one platform is always cheaper.

Can AWS identify spending by developer?

AWS supports IAM principal attribution. Shared-role deployments need suitable identity or session-tag configuration. Telemetry supplies finer session detail.

Does Claude Code’s cost estimate equal my AWS invoice?

No. Its cost metrics are approximate. Reconcile them with AWS billing, accounting for pricing configurations, discounts, and supporting services.

Make Bedrock Costs Accountable

Start with accurate pricing, reliable ownership, and complete reporting coverage. Combine AWS billing data with session telemetry, then add budgets and model policies.

That gives engineering leaders a clearer view of Bedrock Claude cost and the work it supports. Explore more implementation guides on the Opviera blog.