Microsoft Foundry is Microsoft's Azure-based platform for AI models, and Claude is available through it. For a Microsoft-standardized enterprise the logic mirrors the other clouds: Claude arrives through an existing vendor relationship, existing Azure billing, and existing governance, rather than as a new supplier your procurement and security teams must evaluate from scratch.
The resource is the unit of everything
Access to Claude on Foundry is organized around a Foundry resource — the Azure object your team creates, places in a region, and attaches to a subscription and resource group. The resource is where keys are issued, where usage is metered, and where access is administered. In practice, most organizations create separate resources per environment (development, staging, production) so that credentials, quotas, and spend stay cleanly separated. The step-by-step mechanics are covered in Creating and Configuring a Foundry Resource for Claude.
Keys and the SDK
Unlike Bedrock and Vertex, which authenticate with cloud-native credentials, Foundry uses API keys tied to the resource. That is simpler to start with and demands more discipline in production: keys must live in a secret manager, never in code or configuration files, and someone must own rotation. The official Python SDK's Foundry client takes the key and the resource name directly; model IDs are the bare first-party form with no prefix.
from anthropic import AnthropicFoundry
client = AnthropicFoundry(
api_key="<from-your-secret-manager>",
resource="my-foundry-resource",
)
message = client.messages.create(
model="claude-sonnet-5",
max_tokens=1024,
messages=[{"role": "user",
"content": "Extract the renewal date from this clause."}],
)
print(message.content)Feature support: broad, but read the labels
Foundry's coverage of the Claude feature set is wider than Bedrock's or Vertex AI's. The core — Messages API, streaming, tool use, vision, adaptive thinking, prompt caching — is available, as it is on all four platforms. Beyond the core, capabilities such as the Batch API, Files API, code execution tool, and web fetch tool are mostly offered in beta as of July 2026, where Bedrock and Vertex do not offer them at all.
"Beta" is not a synonym for "broken," but it is a synonym for "subject to change." For an enterprise the sensible reading is: build production dependencies on the core feature set freely, and treat beta features as pilot material — useful for proving a concept, risky as a load-bearing part of a system with an SLA to your own customers. If a beta capability is central to your roadmap, note that Claude Platform on AWS offers those features in step with the first-party API, and factor that into platform choice.
Fitting into the Azure estate
Because the Foundry resource is a standard Azure object, your organization's usual machinery applies around it: subscription and resource-group structure for ownership, role-based access for administration, cost views for spend tracking, and your existing audit posture. Compliance works the same way it does on every 3P platform — Claude on Foundry inherits your cloud provider's compliance posture, and the specific certifications and their scope are something to confirm with Microsoft, not something any guide (including this one) can promise you.
Pricing follows the pattern common to the cloud marketplaces: list prices match Anthropic's first-party list prices (for example, Claude Sonnet 5 at $3 per million input tokens and $15 per million output tokens, with introductory pricing of $2/$10 through August 31, 2026). Committed-use discounts, if your volume justifies them, are negotiated with Microsoft as part of your Azure relationship.
Where to go next
Start hands-on with Your First Claude Call on Microsoft Foundry, then work through the resource setup guide to make it production-shaped. To see how Foundry's feature coverage compares with the other three surfaces, check the feature matrix.