Product / UX / UI & API

Azure AI Model Inference

Managed inference service exposing OpenAI models, open-weights models such as Llama, Mistral and Phi, and Microsoft's own models through one OpenAI-compatible REST endpoint. The endpoint switches between model deployments and applies content filtering, rate limiting and keyless authentication. Deployments run either serverless per token or on provisioned dedicated capacity.

The beta Azure AI Inference SDK is deprecated and retires 26 August 2026 in favor of the OpenAI /v1 API; the retirement banner is on the endpoints page. Verified 2026-08-13 via the Microsoft Foundry Models endpoints documentation.

Openness

1 high confidence

1.0

license: proprietary(Azure managed service)
source: closed
interface: OpenAI-compatible REST endpoint over Foundry model deployments
hardware: Azure-managed

Fully proprietary managed Azure service; no source released. Scored closed.

https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/endpoints recorded 2026-08-13
single managed Azure inference endpoint over model deployments; proprietary Azure resource, OpenAI-compatible API

Adoption

4 low confidence

4.0

Default inference surface for Microsoft Foundry across Azure's enterprise base; broad distribution but no published per-service usage number. Level 4 reflects Azure-enterprise scale (reported), not a measured count.

https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/endpoints recorded 2026-08-13
unified endpoint for all Foundry model deployments, enterprise Azure integration

Capability

3 medium confidence

3.0

Capability here is the breadth of a managed endpoint - multi-model routing, authentication, governance - rather than raw engine throughput. It behaves more like a managed API gateway over model deployments than a weight-loading inference engine. The Azure AI Inference beta SDK is deprecated in favour of the OpenAI/v1 API, which changes the client surface rather than what the service does.

https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/endpoints recorded 2026-08-13
single endpoint switches between model deployments with content filtering, rate limiting, keyless auth

Verified 2026-08-13