
Azure OpenAI is Microsoft's service for running OpenAI's models inside Azure. You deploy a model to a resource in your own subscription, call it through an Azure endpoint, and pay for it on your Azure bill. Microsoft Foundry, called Azure AI Foundry until its latest rename, is the platform built around that service. A Foundry resource runs the same Azure OpenAI models. It adds models from other providers, Foundry Agent Service for building agents, evaluations and Foundry Tools such as Speech and Vision. Microsoft's upgrade guide calls the Foundry resource "a superset of capabilities compared to the Azure OpenAI resource type". It also states that after an upgrade "you don't pay more for existing Azure OpenAI functionality."
| Azure OpenAI resource | Foundry resource | |
|---|---|---|
| Models | OpenAI's models | Runs the same Azure OpenAI models |
| Workload | Call Azure OpenAI by API if that is all your workload needs | The work grows past one model: models from other providers, the agent service, evaluations and Foundry Tools |
| Price | Input and output tokens each model processes, at the rates on the Azure OpenAI pricing page | After an upgrade you don't pay more for existing Azure OpenAI functionality |
| Upgrade | You can defer the upgrade or opt out | Keeps its name, endpoint, API keys, network and identity settings |
The other difference from OpenAI's own API is where your prompts go. Microsoft's data, privacy and security page states that prompts and completions are not available to OpenAI and are not used to improve the models. It also states that the models do not interact with services OpenAI operates, such as ChatGPT or the OpenAI API. Prompts are processed in the geography you choose, unless you pick a Global or DataZone deployment type. Data the service stores, such as files uploaded for fine-tuning, stays at rest in the resource in your Azure tenant. Microsoft encrypts it with AES-256 by default.
No. Microsoft's upgrade guide says both resource types "are generally available and supported before and after the upgrade." What changed is the platform's name and the resource that holds your models. Microsoft has renamed the platform twice: Azure AI Studio became Azure AI Foundry, which is now Microsoft Foundry. Microsoft's own table of the change also renames Azure AI Services as Foundry Tools. In the same table, a single Foundry resource with projects replaces a hub, an Azure OpenAI resource and an Azure AI Services resource. Agents move from the Assistants API to the Responses API. Monthly api-version parameters give way to v1 stable routes. The Azure AI User role is now Foundry User, and its role ID is unchanged, as Microsoft's cost guide notes.
Microsoft now upgrades eligible Azure OpenAI resources to Foundry resources itself, in stages. You can defer the upgrade or opt out (Manage automatic upgrades). You can also upgrade a resource yourself and roll it back. Either way, the resource keeps its name, endpoint, API keys, network and identity settings, fine-tuning jobs, batches and stored completions.
You can still call Azure OpenAI by API if that is all your workload needs. The Foundry resource is worth it when the work grows past one model and needs models from other providers, Foundry Agent Service, evaluations or Foundry Tools. Check your role assignments and policies before any upgrade. One written with a wildcard, which does not name a resource kind, can give users the Foundry-only features the moment the resource is upgraded. Microsoft's upgrade guide tells IT administrators to review them first for that reason.
When a team moves from one pilot to several production workloads, we pick the models that fit each workflow. They can come from Azure OpenAI or from the open models in the Foundry Models catalog. We set up the orchestration, the logic that routes work between models and tools, in your own subscription. We test it in staging, on test or masked records, before it goes live.
You are not billed twice when you call Azure OpenAI models through Foundry. Foundry has no price of its own. Microsoft's cost guide notes that it has no page in the Azure pricing calculator "because Foundry is composed of several optional Azure services". A Foundry resource bills the services you use. Azure OpenAI models appear as meters, the line items on your Azure bill, under each Foundry resource. Each model has meters for its input and its output.
On a standard deployment you pay as you go for the input and output tokens each model processes, at the rates on the Azure OpenAI pricing page. Tokens are the small chunks of text a model reads and writes. Provisioned throughput bills differently. You pay by the hour for every provisioned throughput unit (PTU) you deploy, whether or not the deployment is handling requests. An Azure Reservation for a one-month or one-year term lowers that hourly rate (provisioned throughput billing).
The other charges come from what you add. New Foundry capabilities you use, such as Foundry Agent Service, can carry their own pricing. Services such as Azure AI Search and Foundry Tools bill at their own rates. A fine-tuned model bills three ways. Training is billed per token or per hour. Hosting is billed per hour while the model is deployed, even when nobody calls it. Inference, meaning each call to the model, is billed per 1,000 tokens. Azure Prepayment pays for Models sold by Azure, Azure OpenAI included, but not for models billed through Azure Marketplace (Plan and Manage Costs).
Azure OpenAI has no hard spending limit of the kind OpenAI's own API offers. Microsoft's cost guide points to budgets and alerts in Cost Management instead. When we scope a Foundry build, we estimate the running cost of the workflow. We check the usage estimate against the Azure bill before it goes live.
Which setup fits depends on your workload: one model called by API, several models with agents and evaluations, or provisioned throughput for steady volume. A Foundry build with us starts with an audit of your Azure environment and a map of its data connections. We agree the target workflow with your team and settle whether Foundry is the right platform for it before anyone writes code.
Azure OpenAI is Microsoft's service for running OpenAI's models in Azure. You deploy a model to a resource in your own subscription and call it through an Azure endpoint. You pay per token or per provisioned throughput unit, on your Azure bill. Microsoft states that your prompts and completions are not available to OpenAI and are not used to improve the models.
Azure OpenAI is the model service, and Microsoft Foundry is the platform around it. A Foundry resource runs the same Azure OpenAI models. It adds models from other providers, Foundry Agent Service, evaluations and Foundry Tools. Microsoft's upgrade guide calls the Foundry resource a superset of the Azure OpenAI resource. Upgrading keeps the endpoint and API keys you already use.
Yes. Microsoft renamed Azure AI Foundry as Microsoft Foundry, and Azure AI Foundry was itself the new name for Azure AI Studio. Microsoft's own table of the change also renames Azure AI Services as Foundry Tools. Your work moves onto a single Foundry resource with projects, in place of a hub and separate service resources.
Not for what you already run. Microsoft states that after the upgrade you don't pay more for existing Azure OpenAI functionality. The resource keeps its name, endpoint and API keys. New Foundry capabilities you start using, such as Foundry Agent Service, can carry their own pricing. Services such as Azure AI Search bill at their own rates.
No. Foundry has no price of its own. Azure OpenAI models bill as input and output meters under your Foundry resource, at the rates on the Azure OpenAI pricing page. You pay more only for what you add or choose. That includes Azure AI Search, a provisioned throughput deployment, or hosting a fine-tuned model, which bills by the hour while it is deployed.
Yes. If Microsoft has selected your resource for automatic upgrade, you can defer it from the Resource upgrade page in the Azure portal. To opt out, set the upgrade mode to Disabled in your Bicep or Terraform definition. If an upgraded resource causes problems, you can roll it back to Azure OpenAI. To do that, delete its projects, connections and non-OpenAI model deployments first.
Planning your Azure OpenAI or Foundry setup? Book a discovery call. A senior partner runs the call and tells you whether Foundry is the right platform for the job before anything is scoped.