Microsoft has added Anthropic’s Claude 3.5 family to Azure AI Foundry, letting enterprises evaluate and deploy the large language model directly from their Azure portal — without signing up for a separate Anthropic account. The move, effective this week, means Azure handles all billing, authentication, and token management for Claude workloads.

What’s New: Claude Joins the Model Catalog

Azure AI Foundry, formerly known as Azure AI Studio, now lists Anthropic’s Claude 3.5 Sonnet and Claude 3 Haiku in its model catalog. Both are available as serverless APIs, meaning developers pay only for the tokens they consume and never touch the underlying infrastructure. The integration mirrors the recent addition of Mistral and Meta models on Azure, reinforcing Microsoft’s strategy to offer a multi-model playground rather than a walled garden.

The models appear alongside OpenAI’s GPT-4o and the rest of the Azure AI portfolio. Deployment is straightforward: from the Foundry portal, select Claude 3.5 Sonnet, accept the license terms, and within minutes you get an API endpoint and keys tied to your Azure subscription. No separate Anthropic API key or contract is required — billing flows through your existing Azure commitment.

Model Input Price (per 1M tokens) Output Price (per 1M tokens)
Claude 3.5 Sonnet $3.00 $15.00
Claude 3 Haiku $0.25 $1.25

Prices reflect Azure’s published rates as of June 2025 and may vary by region.

What It Means for Azure Users

For enterprise developers and IT architects, the integration removes a significant procurement barrier. Instead of negotiating a separate contract with Anthropic, teams can spin up a Claude endpoint through the same Azure portal they already use for compute, storage, and OpenAI models. Authentication is handled by Microsoft Entra ID (formerly Azure AD), so you can enforce conditional access policies, managed identities, and role-based access control — no shared API keys floating around.

For startups and smaller teams, the serverless pricing model lowers the financial threshold. There’s no base fee or provisioned throughput requirement; you pay only when the model processes a prompt. This mirrors the consumption-based pricing that made Azure OpenAI Service popular.

For Windows developers and sysadmins who live in the Microsoft ecosystem, the integration means Claude becomes a first-class citizen in tools like Visual Studio Code Copilot extensions, Azure Logic Apps, and Power Automate. A customer-service bot running on Windows Server can now call Claude via the same Azure SDK used for OpenAI, with unified logging and monitoring through Azure Monitor.

Data governance is another headline. Azure’s infrastructure provides data residency guarantees and a private networking option. Models deployed as serverless APIs can be hosted in specific Azure regions — currently West US, East US, and West Europe for Claude — giving compliance-conscious organizations a clear data-at-rest location.

How We Got Here

Microsoft’s relationship with Anthropic hasn’t always been cozy. For years, Azure’s AI story revolved almost exclusively around OpenAI, with GPT-4 powering everything from GitHub Copilot to Azure OpenAI Service. But two forces pushed the partnership forward.

First, enterprise demand for choice. Many large Azure customers told Microsoft they wanted a model-agnostic platform. Some were concerned about single-vendor risk with OpenAI, especially after the November 2023 leadership turmoil and subsequent outages. Others simply preferred Claude’s strengths in long-form reasoning, safety filtering, or its 200K token context window — still larger than GPT-4o’s 128K.

Second, competitive pressure from AWS and Google Cloud. Amazon Bedrock gave customers early access to Claude alongside other models; Google’s Vertex AI similarly built a multi-model catalog. Azure needed to keep up, and a serverless Claude offering was the fastest way to signal openness.

Behind the scenes, the technical integration relies on the same Azure Machine Learning serving stack that hosts other third-party models. Anthropic exposes a compatible API, Microsoft’s front-end handles token metering and billing, and the model runs on what Azure describes as “GPU-accelerated infrastructure managed by Anthropic and Microsoft.” The exact hosting arrangement is confidential, but it’s clear that neither company is handing over proprietary weights or training data.

This week’s launch follows a series of preview periods and limited engagements with select customers over the past three months. General availability means Azure support SLAs and Microsoft’s standard service-level agreements now apply to Claude workloads.

How to Get Started

If you already have an Azure subscription with access to Azure AI Foundry, here’s the fastest path:

  1. Navigate to ai.azure.com and sign in with your Azure credentials.
  2. Open the Model Catalog — either from the left navigation or the landing page’s “Explore models” button.
  3. Filter by “Anthropic” or search for “Claude.”
  4. Select Claude 3.5 Sonnet (or Haiku), then click “Deploy.”
  5. Choose the “Serverless API” deployment option, pick a region, and accept the license.
  6. Once deployed, go to the endpoint details pane. Copy the endpoint URL and one of the keys.

From your code, use the standard Azure AI Inference SDK or the Anthropic Python/TypeScript SDKs pointed at your Azure endpoint. The SDK handles authentication; just inject the endpoint and key. Example in Python:

import os
from azure.ai.inference import ChatCompletionsClient
from azure.ai.inference.models import SystemMessage, UserMessage
from azure.core.credentials import AzureKeyCredential

client = ChatCompletionsClient(
    endpoint=os.environ["AZURE_AI_ENDPOINT"],
    credential=AzureKeyCredential(os.environ["AZURE_AI_KEY"])
)

response = client.complete(
    messages=[SystemMessage("You are a helpful assistant."),
              UserMessage("Summarize the latest Windows security updates.")],
    model="claude-3-5-sonnet-20241022"
)
print(response.choices[0].message.content)

Admins can also restrict deployment access using Azure Policy. By creating a policy that allows only specific model IDs, you prevent shadow IT from spinning up unapproved AI services. Microsoft’s documentation provides a ready-to-use policy definition for the “Microsoft.CognitiveServices/accounts” resource provider.

If you’re migrating from direct Anthropic API access, keep an eye on three things:
- Rate limits: Azure imposes default limits per subscription (currently 60 requests per minute for Claude 3.5 Sonnet). You can request an increase through the usual support channel.
- Data logging: By default, Azure may log prompts and completions for abuse detection. Check your content filtering settings if you handle sensitive data.
- Model versioning: Azure pins to a specific snapshot rather than automatically updating. You’ll need to redeploy to get the latest model version.

What to Watch Next

Anthropic has already announced Claude 3.5 Opus, its forthcoming largest model, and hinted at enterprise features like citations and longer output. Once released, expect it to appear in Azure AI Foundry within weeks — the two companies have established a pipeline that shortens the gap from raw model to Azure availability.

More immediately, Microsoft is likely to surface Claude models in places like the Power Platform AI Builder and Microsoft 365 Copilot integrations. While nothing official has been announced, the model-agnostic architecture inside Foundry makes it technically trivial to swap or add models behind the same API surface.

For Windows users and IT pros, the key takeaway is this: Azure is no longer an OpenAI shop. It’s a model supermarket, and Claude just took a prominent shelf. That’s good news for anyone who wants to test, compare, and deploy the best AI for the job — without leaving the Microsoft billing and security umbrella.