Homestead Foundry

Appearance

Available models: Azure AI Foundry ​

Scope: Azure AI Foundry

This page describes the Azure AI Foundry target, the hosted-cloud target of ADR-0011. Foundry Local and Azure Local Foundry differ from it in models, features, identity, cost, and operations. Compare all three on Deployment targets.

This page answers "what can I deploy on Azure AI Foundry." It is the companion to the model catalog, which answers the different question "what has this project chosen, and why." If you are unsure which one you want, read chosen versus available first.

This page does not list every model, on purpose ​

Azure's model catalog is large and it changes continuously. A hand-maintained copy of it in a repository would be wrong within days, and this repository treats a published page that asserts something no longer true as a defect rather than as bookkeeping (see methodology). Copying the catalog here would manufacture exactly that defect.

So this page gives you three things instead: the command that answers the question for your subscription and region, a dated snapshot for scale, and a pointer to Microsoft's own catalog, which is always the authority.

The two on-premises targets are different, and they get full lists, because their catalogs are bounded and enumerable. See Available models: Foundry Local and Azure Local Foundry.

There is now a generated comparison, and it does not contradict this

The reasoning above is about a hand-maintained copy, which would be wrong within days. The model availability matrix is generated from a live read of every region and carries the command that regenerates it, so it fails differently: it goes out of date visibly, on a stated snapshot date, rather than quietly. Use it to compare regions; use the commands below to answer for your own subscription today.

Ask your own subscription ​

Two questions, two commands. Both are read-only.

What can be deployed in a region? This is the one to run when you are choosing a region or checking whether a model has reached you yet.

bash
az cognitiveservices model list --location eastus -o table

Narrow it to a publisher, since the full result is long:

bash
az cognitiveservices model list --location eastus \
  --query "[?model.format=='Microsoft'].{name:model.name, version:model.version, kind:kind}" \
  -o table

What can this account deploy? Run this against a Foundry account you already have. It reflects that account's kind, region, and enabled features, so it is narrower and more accurate than the region query.

bash
az cognitiveservices account list-models \
  --name <your-foundry-account> \
  --resource-group <your-resource-group> \
  -o table

Both commands need only reader access. Neither creates or changes anything.

Snapshot: East US, retrieved 2026-07-30 ​

A point-in-time count, recorded for scale rather than as a list to rely on. It was produced by the first command above, in the region this repository deployed into. Re-run the command rather than trusting these numbers.

az cognitiveservices model list --location eastus returned 258 entries across 12 publishers:

Publisher (model.format)Entries
OpenAI92
Meta38
Microsoft30
Mistral AI20
xAI18
Cohere18
DeepSeek16
Anthropic8
MoonshotAI6
Black Forest Labs6
OpenAI-OSS4
Alibaba2

By account kind: 129 AIServices, 78 MaaS, 46 OpenAI, 5 MAI.

An entry is a model version, not a distinct model, so the count of distinct model names is lower. One model with four published versions is four entries.

Set against that, this project's catalog records roughly 38 rows, of which 9 are deployed. That gap is the point of this page: the catalog is a curated shortlist reflecting one methodology's decisions, not a menu of what Azure offers.

Microsoft's catalog is the authority ​

Keeping this current ​

The snapshot above is hand-run and dated, which makes it the weakest content on this page. Generating it from a live subscription, and generating the same for the two local targets, is tracked as a feature request rather than left as a recurring manual chore: see issue #15, auto-refresh the model availability catalogs.

See also ​