Azure OpenAI (Microsoft Foundry Models)

Azure OpenAI is OpenAI’s model family hosted and operated by Microsoft inside Azure. Microsoft’s docs now place it under Microsoft Foundry: models “sold by Azure” are “hosted by Azure and operated by Azure as part of the Foundry Models service” and “include all Azure OpenAI models and specific, selected models from top providers”. The catalog has two categories: Foundry Models sold by Azure, and Foundry Models from partners and community. Older “foundry-classic” docs paths still exist, so both names appear.

Deployment types and billing model

Names seen on the Microsoft Learn page (updated 2026-10-29): Standard (regional, pay-as-you-go), Global Standard, Data Zone Standard, Developer, Provisioned (provisioned throughput units) and Batch. Some models have separate short-context and long-context pricing categories. Rates are on Microsoft’s Azure OpenAI pricing page and are not recorded here.

Models and fine-tuning

  • The page lists OpenAI GPT-4.1, GPT-4o, o-series, GPT-5 and later, and realtime and audio models, plus open-weight and partner models (for example Llama 3.3 70B, Qwen3, gpt-oss, Mistral, Grok, Cohere and DeepSeek). Model availability changes quickly; check the page.
  • Fine-tuning is supported for several models, with Managed Training (SFT, DPO, RFT) and Programmed Training. The page notes that global training does not offer data residency.

Not confirmed

The overview page itself could not be fetched, so data-retention, privacy, content-filtering and “on your data” (RAG) statements are not recorded. Check Microsoft’s data and privacy page before relying on any of them.

Where it fits

It is the enterprise route to OpenAI models with Azure security and regions. Compare the direct API in ai-api-platforms; pair with azure-ai-search for retrieval and azure-machine-learning for MLOps; see microsoft-foundry-note and microsoft-azure.

Sources