All Services

AI / ML

Amazon Bedrock

Amazon Bedrock is a managed platform for building generative-AI applications with foundation models, agents, knowledge bases, guardrails, evaluation, and model-customization capabilities behind AWS APIs.

Explore pricing models, common use cases, infrastructure support, and the AWS services that commonly work with Amazon Bedrock.

Amazon Bedrock pricing and cost programs

Pricing model: Generative AI model and feature usage

On-Demand
Available
Reserved Instances or reserved capacity
Service-specific
Savings Plans
Not applicable
Spot
Not applicable

Billing dimensions: Input and output tokens · Images or media · Provisioned throughput · Agents and knowledge-base features

Programs and modes: On-Demand inference · Batch inference · Provisioned Throughput · Model customization

Model providers, inference tiers, Regions, and Bedrock features have separate rates.

Free Tier: Service-specific — verify current offers

Pricing reviewed 2026-07-25. Reviewed against the linked official AWS pricing page. Recheck regional rates and program terms before purchase.

Official AWS pricing

Official AWS sources reviewed 2026-07-21.

Why implement Amazon Bedrock?

  • Provides access to multiple foundation-model families without operating GPU inference fleets or separate provider infrastructure.
  • Combines model invocation with managed knowledge bases, agents, guardrails, evaluation, and observability features for production application patterns.
  • Integrates with IAM, KMS, VPC endpoints, CloudWatch, CloudTrail, and other AWS data services for governed enterprise use.

How to implement Amazon Bedrock

  1. Define the task, data classification, quality and safety measures, latency target, Region requirements, and budget before comparing supported models with representative evaluations.
  2. Grant a workload role only the required Bedrock and data actions, configure encryption and private connectivity as needed, and verify the selected model's current data-use and retention terms.
  3. Build the invocation or retrieval workflow with input and output validation, guardrails, bounded retries and timeouts, traceable citations where applicable, and human review for consequential decisions.

Amazon Bedrock best practices

  • Treat model output as probabilistic: test against versioned evaluation sets, monitor drift and failure modes, and never let unverified output silently make high-impact decisions.
  • Minimize sensitive prompt data, defend against prompt injection and data exfiltration, isolate retrieved content by authorization context, and validate tool calls before execution.
  • Record model ID and configuration, token use, latency, guardrail interventions, and application outcomes; set quotas and cost alarms and provide a safe fallback when models or dependencies fail.

Amazon Bedrock use cases and server impact

  • Retrieval-augmented assistants
  • Content generation and summarization
  • Agentic workflows with governed tool use

Replaces self-hosted foundation-model serving for supported models, while prompt design, retrieval authorization, application logic, evaluation, safety controls, and human oversight remain customer responsibilities.

Official implementation resources

Commonly paired AWS services