Deploy a Chat Model in Microsoft Foundry

Lab overview

Give your Foundry project a model to work with. In this lab, you browse the model catalog, deploy a chat model with custom settings as contoso-chat, test it in the playground, and review how the deployment draws from your subscription's quota. The name you choose here is the one every app and agent in the series will point to.

Why you'd do this

A model in the Foundry catalog isn't something you can call. A deployment is: it's a named instance of a model, running in your resource, with its own endpoint, throughput, and guardrails. Every agent, playground, and app in a project points at a deployment name rather than a model name, which is the reason this lab names it contoso-chat instead of something tied to a version. When a newer model ships, you can repoint the deployment and nothing downstream changes.

The deployment type is the decision that matters most on this screen. Global Standard routes requests to whichever Azure region has capacity, which is the cheapest and most available option and fine for a help desk assistant. Data Zone Standard keeps traffic inside a geography such as the EU or US for data-residency requirements, Standard pins it to the resource's own region, and the Provisioned types reserve throughput at a fixed cost for workloads that can't tolerate throttling. Picking the wrong one is expensive or non-compliant, and you can't change it after deployment without redeploying.

The tokens-per-minute rate limit is where quota gets spent. Each subscription has a fixed TPM pool per model and region, and every deployment takes a slice of it. Setting 50,000 TPM here leaves room for the evaluation and agent deployments later in the series; max it out and the next deployment fails with a quota error. The Quota page you visit at the end is the first place to look when a deployment refuses to create.

For certification prep, this maps to the AI-901 objectives on AI agents and their components and the AI-103 material on creating agents in Foundry Agent Service, both of which assume you understand the difference between a model and a deployment.

Lab details

• Model: gpt-5.6-sol
• Deployment Name: contoso-chat
• Deployment Type: Global Standard
• Rate Limit: 50,000 tokens per minute
• Guardrail: Microsoft.DefaultV2

Learning objectives

1. Find and evaluate a model in the Foundry model catalog.
2. Deploy a chat model with a custom name, deployment type, and rate limit.
3. Test a deployment in the chat playground and read its details.
4. Explain how deployments consume subscription quota.

Exam relevance

• AI-901 — Microsoft Azure AI Fundamentals — AI agents and their components, grounding with your own data.
• AI-103 — Developing AI Apps and Agents on Azure — create a prompt agent in Foundry Agent Service, agent instructions, tools, and versions.

How this lab fits into the project

This is lab 2 of 12 in the Contoso Knowledge Assistant project. Contoso's IT help desk fields the same policy questions all day, so over this series you'll build an AI agent that answers them from the company's own documents, then ground it, secure it, evaluate it, monitor it, and publish it to Teams. This lab deploys the chat model, named contoso-chat, that every later lab talks to.

Step-by-step instructions

1. Open the model catalog:
   • Select Discover in the upper-right navigation.
   • Click Models in the left pane.
2. Read the model card:
   • Select gpt-5.6-sol and review its availability, hosting, context window, and pricing.
3. Deploy with custom settings:
   • Click Deploy, then Custom settings.
   • Double-click the default name and change it to contoso-chat.
   • Open the Deployment type dropdown, review the options (Global Standard, Data Zone Standard, Standard, the Provisioned types, and Batch), and select Global Standard.
   • Expand Model version settings and leave the version at the default.
   • Double-click the Tokens per Minute Rate Limit value and type 50000.
   • Leave Guardrails set to DefaultV2 and click Deploy.
   • Wait for the deployment to complete.
4. Verify the deployment:
   • In the playground chat field, type What does an IT helpdesk do? and click the send button.
   • Confirm the model responds.
   • Click the Details tab and review the endpoint, deployment type, and rate limit.
5. Read the quota:
   • Select Manage in the upper-right, then click Quota in the left pane.
   • On the Token per minute tab, select your model's row.
   • In the details pane, review allocated versus total quota and the affiliated deployments list showing contoso-chat.
   • Select the Provisioned throughput unit tab and enable Show all.
   • Note your available quotas and the Request quota button in the upper-right corner.
   • Click Microsoft Foundry to return to the home page.