o4-mini<\/code>.<\/li>\n- Test your model’s chat completions using both a
curl<\/code> command and a Python script.<\/li>\n- Monitor the token usage of your model deployments.<\/li>\n
- Understand the key advantages of using Azure OpenAI over the standard OpenAI API and more..<\/li>\n<\/ul>\n
Lets get started.<\/p>\n
What is Azure OpenAI?<\/h2>\n
Before we get into the hands-on guide, let\u2019s first understand what Azure OpenAI is.<\/p>\n
Azure provides a cloud-based service to access OpenAI’s Large Language Models<\/strong> (LLMs), including GPT-3<\/code>, GPT-4<\/code>, and embedding models. <\/p>\nSince it runs on Microsoft Azure\u2019s cloud infrastructure, you can integrate it with other Azure services, security features, and infrastructure.<\/p>\n
With Azure OpenAI, you can fine-tune OpenAI models like GPT-3.5 using your own data to meet your model requirements. However, this feature is only available in select Azure regions<\/strong>, so availability may vary depending on your location.<\/p>\n\n\ud83d\udca1<\/div>\nFine-tuning means teaching the model with your own examples<\/strong><\/b> so that it gives better or more specific answers for your use case.<\/div>\n<\/div>\nAlso, Microsoft Azure is currently the only cloud provider<\/strong> that offers OpenAI\u2019s models as a fully managed, hosted service through Azure OpenAI.<\/p>\n![\"Azure](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-153.png\")
<\/figure>\nIf you want to understand enterprise use cases for Azure OpenAI, please read this official Azure blog<\/a>.<\/p>\nSetup Prerequisites <\/h2>\n
Here is what we are going to do in high level.<\/p>\n
\n- Azure OpenAI resource creation <\/li>\n
- Deploy model using Azure Foundry<\/li>\n<\/ol>\n
For this, <\/p>\n
Ensure you are a privileged user (e.g., with Contributor<\/code> role) or owner on the Azure cloud to create and manage the Azure OpenAI service.<\/p>\nAlso, you must have access to the Azure portal (UI)<\/p>\n
Create Azure OpenAI Resource<\/h2>\n
The first step is to create a Azure OpenAI resource.<\/p>\n
The resource serves as your gateway to OpenAI models through Azure’s infrastructure. It provides the endpoint URL and authentication keys<\/strong> needed to make API calls.<\/p>\nFollow the steps below to create the OpenAI service.<\/p>\n
In the search bar, type Azure OpenAI<\/code> and select the OpenAI service.<\/p>\n![\"typing](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/06\/image-45.png\")
<\/figure>\nClick the + Create<\/code> button to create a new OpenAI resource.<\/p>\n![\"creating](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/06\/image-46.png\")
<\/figure>\nOn the Basics<\/code> page, select the Subscription, Resource Group name (create one if you dont have one), Region, Name, and the Pricing tier. I am going with Standard (On-Demand) pricing tier.<\/p>\n![\"Azure](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/06\/image-62.png\")
<\/figure>\nOn the Network<\/strong> page, you can control who can access your Azure OpenAI resource. For this demo, I’m choosing “All networks”<\/strong> to keep it simple and accessible from anywhere.<\/p>\n\n\ud83d\udca1<\/div>\nFor a production setup<\/strong><\/b>, it’s better to choose a more secure option<\/strong><\/b>, like:<\/p>\n1. Selected networks<\/strong><\/b> \u2013 to allow access only from specific IPs or virtual networks.<\/p>\n2. Private endpoints<\/strong><\/b> \u2013 for the highest security, where access is only allowed through your private Azure network.<\/p>\nChoose the option that fits your use case and security needs.<\/p><\/div>\n<\/div>\n![\"the](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/06\/image-48.png\")
<\/figure>\nOn the Tags<\/code> page, you can add the necessary tags, and on the Review + submit<\/code> page, you can see the preview of the configuration.<\/p>\nOnce verification is completed, click the Create<\/code> button to create the OpenAI resource. It will take a minute or two for the deployment to be ready.<\/p>\nOnce the deployment is successfully completed, you can go to the overview page of the resource.<\/p>\n![\"the](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/06\/image-57.png\")
<\/figure>\nIn the next page you will find options to get the generic API endpoint and keys under the develop option.<\/p>\n
For example, the endpoint typically looks like:<\/p>\n
https:\/\/<resource-name>.openai.azure.com\/<\/code><\/pre>\n![\"\"](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-60.png\")
<\/figure>\nNow that the resource is created, you need to deploy the specific models you want to use with the endpoint.<\/p>\n
\n\ud83d\udca1<\/div>\nSo when you are creating a resource, you are essentially creating a dedicated instance of the Azure OpenAI service within your Azure subscription. It acts as a logical boundary that defines the scope, permissions, and billing for your AI services.<\/div>\n<\/div>\nDeploying a Model<\/h2>\n
The next step is to deploy a model that we need to use with the created API endpoint.<\/p>\n
To do this, head over to Azure AI Foundry portal<\/a>, which is a dedicated page to manage the Azure AI services.<\/p>\n![\"\"](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-154.png\")
<\/figure>\nNext, you may be asked to log in and then provide the same email address you used to sign in to the Azure cloud.<\/p>\n
Then choose the Deployments option from the left pane.<\/p>\n![\"\"](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-61.png\")
<\/figure>\nClick Deploy Model<\/strong> option and select the Deploy base model.<\/strong><\/p>\n![\"\"](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-62.png\")
<\/figure>\nYou will be prompted to select the model. You can search for an OpenAI model in the search box as shown below. Here I have selected o4-mini<\/code> (Great for fast, low-cost tasks. Good balance of speed and intelligence)<\/p>\nAfter selecting a model, click confirm<\/strong>.<\/p>\n![\"\"](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-63.png\")
<\/figure>\nIn a few seconds, you’ll be prompted to deploy the model. In the popup, you\u2019ll find key deployment settings like deployment type, model version, upgrade policy, token rate limits, and content filters.<\/p>\n
The default Deployment type is Global Standard<\/strong>, meaning the model runs on Azure’s global infrastructure.<\/p>\nThe model version upgrade policy lets you decide when to switch to newer versions<\/strong>, either on release or expiry of the current one. You can also choose a specific model version if needed.<\/p>\nThe token-per-minute rate limit controls how many tokens can be processed per minute<\/strong> for both input and output. Content filters help block harmful or inappropriate content in queries and responses.<\/p>\nNow go ahead and deploy it.<\/p>\n![\"\"](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-67.png\")
<\/figure>\nOnce the deployment is complete, you’ll see options to test the model using the playground, API endpoint, and keys<\/strong>. You’ll also get code snippets for Python, JavaScript, and other languages, as shown below.<\/p>\n![\"\"](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-65.png\")
<\/figure>\n<\/p>\nIf you open the playground, you can interact with the model directly and adjust parameters like token limits and reasoning depth. It also includes a token counter that tracks API usage, helping you monitor both consumption and costs.<\/p>\n![\"\"](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-66.png\")
<\/figure>\nIn the same way, you can deploy additional models based on your requirements and access them using their respective API endpoints and tokens.<\/p>\n
You can check out all the available models from the model catalogue<\/a>.<\/p>\nNow that we’ve validated the model, let’s go over how to use the model’s API endpoint along with the API token to test it.<\/p>\n
How to use the Azure OpenAI APIs?<\/h2>\n
In a real-world scenario, we typically use the model’s API endpoints within our applications, like chatbots or Retrieval-Augmented Generation (RAG) systems. As mentioned earlier, under the deployment section, you can find the model’s API endpoint and your API key.<\/p>\n![\"\"](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-69.png\")
<\/figure>\nAPI Call Using curl<\/h3>\n
First lets test the API call using curl command.<\/p>\n
To start, store your API key as an environment variable on your local machine:<\/p>\n
export AZURE_OPENAI_API_KEY=<YOUR API KEY><\/code><\/pre>\nTo make an API call, you\u2019ll need the following details:<\/p>\n
\n- OpenAI Resource Name (e.g., dcube-models)<\/li>\n
- Deployment Name (e.g., o4-mini)<\/li>\n
- API Version (e.g., 2024-12-01-preview)<\/li>\n<\/ul>\n
The following CURL command sends a request to Azure OpenAI model deployment to generate a response in a chat format.<\/p>\n
curl https:\/\/<OPENAI RESOURCE NAME>.openai.azure.com\/openai\/deployments\/<DEPLOYMENT NAME>\/chat\/completions\\?api-version\\=<VERSION NAME> \\\n -H \"Content-Type: application\/json\" \\\n -H \"api-key: ${AZURE_OPENAI_API_KEY}\" \\\n -d '{\n \"messages\": [\n {\"role\": \"system\", \"content\": \"You are a DevOps Engineer.\"},\n {\"role\": \"user\", \"content\": \"Just need the exact command to perform telnet using curl.\"}\n ],\n \"max_completion_tokens\": 1000\n }' | jq\n<\/code><\/pre>\n\"max_completion_tokens\": 1000<\/code> sets the upper limit on the number of tokens (words\/parts of words) the model can use in its response.<\/p>\nYou\u2019ll get an output similar to the one below. Look for the content<\/code> section where the model\u2019s actual response to your prompt will appear. Here\u2019s a trimmed version showing just that part:<\/p>\n.\n.\n.\n \"finish_reason\": \"stop\",\n \"index\": 0,\n \"logprobs\": null,\n \"message\": {\n \"annotations\": [],\n \"content\": \"You can use curl\u2019s built-in TELNET support. For example, to \u201ctelnet\u201d to host 192.168.1.10 on port 23:\\n\\ncurl -v telnet:\/\/192.168.1.10:23\\n\\nExplanation: \\n- -v : verbose (shows the negotiation and connection info) \\n- telnet:\/\/\u2026 : tells curl to use the TELNET protocol \",\n \"refusal\": null,\n \"role\": \"assistant\"\n.\n.\n.\n<\/code><\/pre>\nAPI Call Using Python<\/h3>\n
If you want to test Azure OpenAI API calls programmatically, you can try the Python example below.<\/p>\n
I am using the same sample code provided by Azure on the deployment details page.<\/p>\n
Create a Requirements file for the Packages<\/p>\n
vim requirements.txt\n<\/code><\/pre>\nAdd the following contents inside the file<\/p>\n
openai~=1.60.2\n<\/code><\/pre>\nBefore installing the packages, create a venv.<\/p>\n
python3 -m venv venv\nsource venv\/bin\/activate<\/code><\/pre>\nInstall the Packages<\/p>\n
pip install -r requirements.txt<\/code><\/pre>\nCreate a Python script to query the OpenAI deployment via API<\/p>\n
vim main.py<\/code><\/pre>\nAdd the following contents. Replace <your-endpoint><\/code> with your model endpoint.<\/p>\nimport os\nfrom openai import AzureOpenAI\n\nendpoint = \"<your-endpoint>\"\nmodel_name = \"o4-mini\"\ndeployment = \"o4-mini\"\n\nsubscription_key = os.getenv(\"AZURE_OPENAI_API_KEY\")\napi_version = \"2024-12-01-preview\"\n\nclient = AzureOpenAI(\n api_version=api_version,\n azure_endpoint=endpoint,\n api_key=subscription_key,\n)\n\nresponse = client.chat.completions.create(\n messages=[\n {\n \"role\": \"system\",\n \"content\": \"You are a senior DevOps engineer helping beginners.\",\n },\n {\n \"role\": \"user\",\n \"content\": \"How do I set up CI\/CD using GitHub Actions for a Node.js app?\",\n }\n ],\n max_completion_tokens=100000,\n model=deployment\n)\n\nprint(response.choices[0].message.content)<\/code><\/pre>\nIn the above code we are loading the subscription_key<\/code> from an environment variable instead of hardcoding it. So you need to export the key using the following. Replace <YOUR-KEY> with your subscription key.<\/p>\nexport AZURE_OPENAI_API_KEY=<YOUR-KEY><\/code><\/pre>\nTo run the code, use the following command.<\/p>\n
python3 main.py<\/code><\/pre>\nOn a successful execution you will get an output like the following.<\/p>\n
$ python3 main.py \n\nHere\u2019s a complete walk-through for adding CI\/CD to your Node.js app with GitHub Actions. We\u2019ll cover:\n\n 1. Prerequisites \n 2. Writing basic tests \n 3. Creating your workflow file (`.github\/workflows\/ci-cd.yml`) \n 4. Caching `node_modules` \n 5. Running lint & tests (CI) \n 6. Deploying (CD) \u2013 example to Heroku \n\n.\n.\n.<\/code><\/pre>\n\n\ud83d\udccc<\/div>\nCheckout the python for devops<\/a> blog for more AI\/ML based use cases for DevOps<\/div>\n<\/div>\nMonitoring Token Usage<\/h2>\n
One of the best ways to optimize the Azure OpenAI usage is to monitor the resource.<\/p>\n
There is a dedicated monitoring page to track the token usage<\/strong> and the performance of the model deployment.<\/p>\nTo get started, navigate to the Monitoring<\/code> section and select the deployment to see the graph based on the metrics.<\/p>\n![\"\"](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-70.png\")
<\/figure>\nNow, if you select the model deployment, it\u2019ll show the total number of requests and how many tokens have been used.<\/p>\n![\"\"](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-71.png\")
<\/figure>\nIn the screenshot above, you\u2019ll see the following token details.<\/p>\n
\n- Prompt token count<\/strong>: This is the number of tokens in the text you send to the model.<\/li>\n
- Completion token count<\/strong>: These are the tokens in the model\u2019s response.<\/li>\n
- Total token count<\/strong>: This is just the sum of prompt and completion tokens.<\/li>\n<\/ul>\n
Tokens are the basic units that language models use to process text. They represent pieces of words<\/strong>, whole words, or even parts of words. The following image illustrates how text gets broken<\/strong> down into tokens by language models.<\/p>\n![\"illustration](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-74.png\")
<\/figure>\nCheckout OpenAI Tokenizer<\/a> to know more.<\/p>\nConfiguring Security and Compliance <\/h2>\n
Protecting data in enterprise environments is very important, and the service must follow all required rules and regulations.<\/p>\n
By default, Azure OpenAI comes with a built-in content filter, but you can customize it using Guardrails + Controls<\/strong> to better fit your specific needs.<\/p>\nTo get started, navigate to the Guardrails + Controls<\/code> for the security configurations.<\/p>\n![\"The](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/06\/image-76.png\")
<\/figure>\nGo to the Try it out<\/code> section to see the available content filtering for text and image, though we can customize the filtering as well.<\/p>\n![\"availabe](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/06\/image-77.png\")
<\/figure>\nWe can even set the level of the content filtering to low to high based on our use case.<\/p>\n
Azure OpenAI Pricing<\/h2>\n
You can navigate to the Model catalog<\/code> , and select the model to see the complete details of that model.<\/p>\n![\"The](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-85.png\")
<\/figure>\nThen from details, click the See direct from Azure pricing<\/code><\/p>\n![\"The](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-86.png\")
<\/figure>\nOn the next page, you will get the link to open the Azure OpenAI Service pricing page.<\/p>\n![\"page](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-87.png\")
<\/figure>\nClicking it will take you to another page where you can see the pricing of the model.<\/p>\n![\"the](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/06\/image-81.png\")
<\/figure>\nNot only this, we can see the region wise price and the fine tuning price as well on this page.<\/p>\n
So how do we calculate OpenAI cost for text generation?<\/p>\n
Features of Azure OpenAI Service<\/h2>\n
Azure offers a couple of features to manage and secure the OpenAI<\/p>\n
\n- Fine Tune Models:<\/strong> We can select and use the trained models, though we can even train those models with our datasets for our use case.<\/li>\n
- Content Safety and Filtering:<\/strong> We can restrict the input and output of the model to avoid the harmful queries and outputs, or even the content that violates the organization’s compliance.<\/li>\n
- Scalability and Availability:<\/strong> Azure OpenAI will automatically scale based on the workload requirements.<\/li>\n
- Batch Job:<\/strong> Azure OpenAI can process batch jobs. For example, if we want to analyze thousands of user reviews, we can create this as a batch job, so that OpenAI will process all the requests and give the complete output.<\/li>\n<\/ol>\n\n\ud83d\udca1<\/div>\nBatch jobs should be in JSON format and should be less than 200MB<\/div>\n<\/div>\n
\n- We can even set the Provisioned Throughput Units (PTUs), meaning the processing capacity, if we already know the workload volume.<\/li>\n
- Azure OpenAI is a region based service, so we can setup in the nearest region for the highest availability.<\/li>\n<\/ol>\n
Azure OpenAI Best Practices <\/h2>\n
The following are some of the best practices.<\/p>\n
\n- Choose the right model based on your use case, because it has cost-effective models (like o4-mini)<\/li>\n
- Optimize the token usag<\/strong>e using the prompt engineering, so that we can reduce the cost.<\/li>\n
- Store the API key in a secure vault like Azure Key Vault<\/strong> and rotate the keys periodically to avoid accidental exposure.<\/li>\n
- Set the token limit so that we can avoid unnecessary overusage.<\/li>\n
- Monitor and set alerts for the usage so that we can track the cost and performance.<\/li>\n<\/ol>\n
Azure OpenAI vs OpenAI API <\/h2>\n
So, how is Azure OpenAI different from OpenAI’s direct API?<\/p>\n
Azure OpenAI runs on Microsoft Azure\u2019s secure cloud infrastructure<\/strong>, with private endpoints, VNets, enterprise-grade compliance and enterprise-level uptime commitments.<\/p>\nWhile the OpenAI<\/a> API operates on OpenAI\u2019s own servers based in the US.<\/p>\nAzure OpenAI has native integration with Azure services (Cognitive Services, Logic Apps, Power Platform etc)<\/p>\n
This makes Azure OpenAI a better fit for enterprise environments, especially when data location and compliance are key concerns. Organizations can keep data within their selected Azure region, which helps meet regulatory requirements<\/strong> like GDPR<\/a> or HIPAA<\/a>.<\/p>\nIn contrast, the OpenAI API might not align with stricter enterprise policies around data handling and compliance.<\/p>\n
Cleaning Up Your Resources<\/h2>\n
It’s really important to clean up any resources you’re not using.<\/p>\n
Once you’re done learning and no longer need the model deployment, go to the deployment page and delete it.<\/p>\n![\"\"](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-77.png\")
<\/figure>\nAfter you delete the deployments, you can also delete the resource you created in Azure OpenAI.<\/p>\n![\"\"](\"https:\/\/storage.ghost.io\/c\/5f\/2f\/5f2f4d20-2abf-4534-8d40-7aa233aedd43\/content\/images\/2025\/07\/image-78.png\")
<\/figure>\nConclusion<\/h2>\n
Now that you\u2019ve set up Azure OpenAI, deployed a model, tested it with curl and Python, and checked out how token usage works, you\u2019ve covered the essentials.<\/p>\n
Also, you also got a glimpse of how Azure takes care of security and compliance<\/strong> right out of the box.<\/p>\nHere is what you should do next.<\/p>\n
Start building<\/strong> chatbots, internal tools, or DevOps assistants using the AI features.<\/p>\nAnd this is just the beginning. Keep exploring, test out different models, and watch for upcoming guides on fine-tuning<\/strong> and getting things ready for production.<\/p>\n
\n