Microsoft's unified GenAI studio, model catalog, and agent runtime
Azure AI Foundry is Microsoft's 2026 re-architecture of the Azure AI stack into a single studio: model catalog (OpenAI, Anthropic, Meta, Mistral, DeepSeek, Cohere, plus Microsoft's own), agent service with hosted agents and tools, evaluation harness, fine-tuning, observability, and content-safety filters in one pane. It replaced the older Azure AI Studio and consolidates what used to be three consoles.
Who it's for: Enterprise platform teams already on Azure, ops teams migrating from Azure OpenAI Service, and any org that needs Anthropic + OpenAI + open-weight under one compliance umbrella.
OpenAI, Anthropic Claude, Meta Llama, Mistral, DeepSeek, Stable Diffusion, plus Microsoft Phi — all deployable from one UI with unified billing.
Agent Service with managed threads, code interpreter, file search, and SharePoint grounding — no LangChain host to wire up.
Pre-registered quality and safety evaluators (groundedness, fluency, harm, jailbreak) runnable as CI checks.
CMK, private endpoints, customer lockbox, EU data boundary — works out of the box for regulated industries.
The most complete "all the models, all the compliance" control plane available in 2026 — heavier and pricier than cobbling Bedrock + a vector DB + LangSmith, but worth it if your InfoSec team reads the Azure bill.