Deploy models to endpoints
Unit 4 of 7 · Module 2: Select, deploy, and evaluate Microsoft Foundry models
Before you start
Short explainers for ideas this lesson builds on. Watch these first if they're new to you.
Go deeper
Unlocks when you've watched the lesson.
Read this lesson
Into service
Once you've picked a model, you deploy it, so your applications can reach it through an endpoint. The Foundry portal walks you through deploying, and gives you tools to test the model right away. First, the deployment type. Each type differs in where your data is processed, how it scales, and how you're billed. Global deployments can use any Azure region. Global Standard is pay-per-token, best for general workloads, with the highest quota. Global Provisioned uses reserved provisioned throughput units, or PT Us, for predictable, high throughput. And Global Batch runs large asynchronous jobs within twenty-four hours, at a fifty percent discount.
Zones and regions
Data Zone deployments keep your data inside a specific data zone, like the EU or the US. Data Zone Standard is pay-per-token, for when you need data zone compliance. Data Zone Provisioned gives predictable throughput with reserved PT Us. And Data Zone Batch handles large asynchronous jobs within the zone. Standard deployments keep processing within your chosen Azure geography, pay-per-token: great for data residency, or low-volume scenarios. Regional Provisioned reserves PT Us within that geography. And Developer deployments, pay-per-token in any region, are only for evaluating fine-tuned models. Each model lists the types it supports, and the portal picks the best option for you. Whenever you can, use Global Standard in a Foundry resource, for the most capabilities.
Deploy it
To deploy, select Discover, then Models, and open the model card to review its specs and supported deployment types. Select Deploy. Choose default settings to go fast with the recommended configuration, or custom settings to change them. If the model needs an Azure Marketplace subscription, which is common for partner and community models, you'll see its terms of use. Review them, and select Agree and Proceed. Models sold directly by Azure, like GPT five mini, don't need one. The deployment name defaults to the model name. You can change it, which helps when you deploy the same model more than once, and your code uses this name in the model parameter to route requests. The portal picks the deployment type, based on the model and your environment. With managed compute, the model runs on dedicated virtual machines billed by the hour, and you choose how many instances run, for load distribution and redundancy. Select Deploy, and when it's done, you land in the playground. Check that the status shows Succeeded.
Manage and test
To manage your deployments, select Build, then Models, to see every deployment in your resource. Open one to see its configuration and status, its endpoint URL, its keys or tokens, monitoring and usage metrics, and options to adjust its settings or delete it. Those details are what your apps need to connect. In the playground, you can test a deployment without writing code. Try simple questions, multi-step reasoning, requests for a specific format or style, and edge cases that might reveal its limits. Use a system message to set the context, tone, and instructions for every input, like: respond as a customer service representative. Tune the parameters: temperature, for creativity versus consistency; max tokens, to limit response length; and top-p, for nucleus sampling. And the Code tab shows how to call your deployment in Python, C sharp, or JavaScript, ready to copy.
Connect your app
To call the model from your application, you need three things from the deployment details. First, the endpoint URL, where requests go. Foundry offers project endpoints for Foundry-specific features, and OpenAI v1 endpoints for broad compatibility with OpenAI APIs. Second, authentication: either a secret key, or, better for production, Microsoft Entra ID, where your app presents a token based on its own identity. And third, the deployment name, used in the model parameter to route each request to your deployment. The portal links SDKs and REST API docs, with samples for requests, authentication, and responses. With your model deployed and tested, you're ready to build it into your app, or move on to deeper evaluation.
Facts from Microsoft's documentation: Understanding deployment types in Microsoft Foundry Models; Microsoft Foundry Models overview; Endpoints for Microsoft Foundry Models (CC BY 4.0), rewritten in Loop's own words and examples. Narration, animation, and quiz by Loop. The coding lab is based on code samples in Microsoft's Azure AI documentation (MicrosoftDocs/azure-ai-docs, MIT License; notice). Microsoft, Azure, and Microsoft Foundry are trademarks of Microsoft. Loop is independent and not affiliated with or endorsed by Microsoft.