Azure OpenAI Service — AI Glossary

Reference details

Related terms

• OpenAI API
• AWS Bedrock
• Google Vertex AI

How it works

1. A named deployment of a specific model version is created inside a cloud resource, and requests address that deployment rather than a global model name.
2. That indirection pins the version: a provider-side model update does not change what the deployment serves until it is redeployed.
3. Authentication uses cloud identity rather than a provider API key, so access is granted by role.
4. Quota is allocated to the deployment per region, which means capacity planning is per deployment rather than per account.
5. The request and response shapes largely match the provider's own API, so porting is mostly endpoint, auth and deployment naming.

Example: The pinning is the feature

A global endpoint silently moves to a newer model snapshot; prompts tuned against the old behaviour drift, and nothing in the application changed to explain it. A named deployment of a specific version continues serving that version until someone redeploys, so the upgrade is a dated action with a rollback. The cost is that staying current is now manual work. Which of those two is the feature depends entirely on whether behaviour stability or newest capability matters more.

Commonly confused with

The same models as OpenAI's own API, reached through a different contract, endpoint shape and deployment model — you deploy a named model into a resource rather than calling a global endpoint. Choose it for enterprise agreements, regional data residency and compliance, not for capability. Version availability trails the direct API, so a model released upstream is not immediately callable here.

When to use it

Reach for it when:

• Enterprise requirements — private networking, residency, cloud identity — that a direct API cannot satisfy.
• Model version must be pinned so provider-side updates cannot change behaviour unannounced.
• The organisation's governance and billing already run through that cloud.

Reach for something else when:

• Newest model access matters most, where the direct API leads.
• A prototype, where deployments and quota requests add friction for nothing.
• Assuming code portability is free — auth, endpoints and deployment naming all differ.

Continue your research

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

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