Amazon Bedrock
Bedrock
You need managed access to foundation models and associated features for a generative AI application.
Reach for it when
- Invoking a suitable model without operating its full inference infrastructure.
- Building retrieval and tool workflows with explicit evaluation and access controls.
Do not reach for it when
- Assuming fluent output is correct or that a managed guardrail eliminates every failure.
- Choosing model access before checking regional availability and task requirements.
Alternatives, and how to choose
| Service | Pick it instead when |
|---|---|
| SageMaker | Choose when custom training or hosting control is central to the workload. |
| Other model APIs | Compare actual task quality, latency, data handling and total cost. |
How you pay
- The model
- Pricing varies by model, inference mode and enabled retrieval, guardrail or other features.
- The line item that surprises people
- Long inputs, repeated tool calls and unconstrained retries can multiply usage charges.
What trips people up
- Model availability, context limits and supported features differ across models and Regions.
- Treat retrieved content and tool responses as untrusted inputs; enforce sensitive actions and permissions outside the model.
Verify the live service
This page is a concept reference. Cost models are qualitative; confirm the current offering, Region and pricing before deploying.