AWS Bedrock — AI Glossary

Reference details

Related terms

• Azure OpenAI Service
• Google Vertex AI
• Anthropic API

How it works

1. Models from several vendors are exposed through one cloud API, so switching vendor is a model identifier rather than a new integration and contract.
2. Authentication and authorisation use the cloud's own identity system, letting model access be granted by role alongside other infrastructure permissions.
3. Requests stay inside the customer's cloud account, which is the data-governance property that usually drives the choice.
4. Throughput is either on-demand or purchased as provisioned capacity, and the latter is what makes latency predictable under load.
5. Managed layers sit above inference — retrieval, agents, content filtering — so a pipeline can be assembled without separate services.

Example: When provisioned capacity is the cheaper option

A workload needs a guaranteed 200,000 tokens per minute at peak. On-demand, bursts hit throttling and the application retries, so effective latency rises exactly when traffic does. Provisioned capacity reserves the throughput and is billed whether used or not — so it is cheaper per token only above a utilisation threshold. Below it, on-demand with backoff wins. The decision is the ratio of peak to average: a workload at 20% average utilisation pays five times the effective rate to reserve its peak.

Commonly confused with

A managed gateway to models from several vendors inside one cloud, which makes it comparable with Azure's OpenAI service and Google's Vertex rather than with the model providers themselves. Two consequences follow: the model you reach is the vendor's, but the contract, region, billing and compliance posture are the cloud's — and feature parity with the vendor's own API commonly lags, so a capability available direct may not be available here yet.

When to use it

Reach for it when:

• The workload runs in that cloud and should inherit its identity, networking and billing.
• Several vendors' models are needed behind one contract and one integration.
• Data must remain inside the customer's cloud account.

Reach for something else when:

• A prototype needing one call, where the setup cost dominates.
• The newest model version matters most, since managed surfaces often lag the provider's own API.
• Deliberate cloud independence, which a cloud-specific gateway works against.

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to AWS Bedrock — AI Glossary.

• Browse models
• Review providers
• Inspect benchmarks
• Explore workflows
• Learn AI concepts
• Read decision guides
• Review trust and methodology
• Start a decision path