No free plan. The cheapest way in is On-Demand (Pay-as-you-go) at From $0.80/1M input tokens (Haiku 3.5) to $15.00/1M input tokens (Opus 4); output tokens billed separately at higher rates (e.g., $4.00/1M output for Haiku 3.5, $15.00/1M output for Sonnet 4, $75.00/1M output for Opus 4). Consider free alternatives in the ai models category if budget is tight.
Functionally you get the same underlying Claude models, but on Bedrock the traffic is authenticated via AWS IAM and SigV4, data stays within your VPC, and billing consolidates onto your AWS invoice. Bedrock also adds managed services (Knowledge Bases, Agents, Guardrails) that have no direct equivalent on Anthropic's API.
Yes. Amazon Bedrock is a HIPAA-eligible service under an executed BAA with AWS and carries SOC 1/2/3, ISO 27001, and PCI DSS compliance. In GovCloud regions it also holds FedRAMP High authorization.
Bedrock bills Claude on a per-model, pay-as-you-go basis using separate input-token and output-token rates. For example, Haiku 3.5 starts at $0.80 per 1M input tokens, Sonnet 4 at $3.00, and Opus 4 at $15.00. Output tokens are billed at higher rates. Provisioned Throughput plans offer committed capacity at hourly rates, and enterprise agreements can negotiate further discounts.
The Claude model families — including Opus 4, Sonnet 4, and Haiku 3.5 tiers — are available on Bedrock, though availability may vary by region. Check the Bedrock console for current model availability in your target region.
Yes. You can create VPC interface endpoints for Bedrock using AWS PrivateLink so that all InvokeModel traffic stays within your private network and never traverses the public internet.
See Anthropic Claude on AWS Bedrock plans and find the right tier for your needs.
See Pricing Plans →Still not sure? Read our full verdict →
Last verified March 2026