Claude Models Land in Microsoft Foundry Alongside OpenAI

Anthropic’s Claude models landed in Microsoft Foundry this week, opening a striking new lane in the multi-cloud AI landscape: Claude in Microsoft Foundry sits next to OpenAI’s GPT family on the same enterprise AI platform, accessible through the same APIs, billed through the same Azure pipeline. The launch breaks Microsoft’s three-year exclusive bias toward OpenAI on its developer-facing AI infrastructure and gives enterprise Azure customers a real choice between Anthropic and OpenAI without leaving the platform. For Microsoft, it’s a multi-model strategy that mirrors AWS Bedrock’s model-agnostic positioning. For Anthropic, it’s distribution into a customer base that previously had to leave Azure to use Claude. For enterprises building on Azure, it’s the option to pick the model that fits the task without architectural rewrites.

What’s actually new

Claude in Microsoft Foundry adds Claude’s full model family (Opus, Sonnet, Haiku variants of Claude 4.5 and earlier generations) to Foundry’s catalog of available models. Foundry — Microsoft’s Azure-native AI development and deployment platform — already supported OpenAI’s GPT family, Meta’s Llama, Mistral’s models, and various smaller models. Adding Anthropic’s Claude completes the trio of major frontier AI providers within a single platform.

The integration is more than catalog addition. Claude models in Foundry expose the same APIs as other Foundry-hosted models: a unified inference endpoint, integrated identity via Azure Active Directory, native Azure billing, and the standard Foundry observability stack. For Azure-native applications, calling Claude is now as straightforward as calling GPT — the developer experience is mostly identical, with model name being the variable. Migration between Claude and GPT for specific workloads becomes a configuration change rather than an architectural one.

The procurement story matters too. Enterprise Azure customers with Enterprise Discount Programs (EDPs) can now route Claude spending through their Azure commitments. The Foundry pricing for Claude largely mirrors Anthropic’s direct API pricing, but with the procurement convenience of consolidated Azure billing. For organizations that have standardized on Azure for AI infrastructure, this removes the procurement friction that previously made multi-provider experimentation slow.

Why Claude in Microsoft Foundry matters

• Microsoft’s multi-model strategy crystallizes. Microsoft has invested heavily in OpenAI and continues to. But adding Claude — OpenAI’s principal competitor — to Foundry is the clearest signal yet that Microsoft is hedging the relationship and prioritizing customer choice over single-provider lock-in.
• Enterprise Azure customers gain real provider choice. Previously, Azure-native applications wanting to use Claude had to integrate Anthropic’s API directly, complicating procurement, authentication, and monitoring. The Foundry integration eliminates the friction.
• The Anthropic distribution story expands meaningfully. Claude is now natively available on AWS (Bedrock + Claude Platform on AWS) and Microsoft Foundry, plus Google Cloud’s Vertex AI. The cross-cloud reach is now comparable to OpenAI’s distribution.
• The cloud-AI alignment story stays nuanced. Microsoft is “OpenAI-aligned” in the strategic narrative but multi-provider in practice. AWS is “Anthropic-aligned” but offers OpenAI via Bedrock. Google Cloud is “Gemini-first” with limited third-party AI. The simplistic alignment narratives miss the multi-provider reality.
• For developers, the productivity story is real. Switching between Claude and GPT for specific tasks (Claude for long-context summarization, GPT for specific reasoning patterns, Haiku for high-volume cheap inference) becomes a config-line change instead of a service-integration project.
• Strategic implications for OpenAI. Microsoft’s largest distribution channel for AI is now multi-provider, which changes the OpenAI competitive context. The previous “Azure customers default to OpenAI” dynamic erodes; OpenAI now competes on merits inside Foundry rather than benefitting from default-vendor positioning.

How to use Claude in Microsoft Foundry today

Claude in Microsoft Foundry is available now for Azure customers in supported regions. The integration uses Foundry’s standard provisioning flow.

1. Confirm your Azure subscription has Foundry access. Open Azure AI Foundry in the Azure portal; verify your subscription appears in the resource list. If not, your admin must enable Foundry for the subscription.
2. Browse the Foundry model catalog. Models from Anthropic should appear alongside those from OpenAI, Meta, Mistral, and others. The Claude family includes Opus, Sonnet, and Haiku variants.
3. Deploy a Claude model. Foundry uses a deployment-based pricing model: deploy the model to a specific resource, then call the deployment endpoint. The deployment process is point-and-click in the Azure portal or scriptable via Azure CLI.
4. Call the model via the Foundry API. The endpoint uses Foundry’s unified inference API, which closely mirrors Anthropic’s native Messages API. Authentication uses Azure AD credentials; billing flows through your Azure subscription.
5. Wire observability. Foundry’s built-in tracing captures every model call with full request/response data, latency, token usage, and cost. Use it directly or export to your existing observability stack via Azure Monitor.

# Sample Python code calling Claude in Microsoft Foundry

# Install the Azure AI Foundry SDK
pip install azure-ai-foundry

# Authenticate via Azure CLI (standard Azure auth chain)
az login

# Example call
from azure.ai.foundry import FoundryClient

client = FoundryClient(
    endpoint="https://your-foundry.eastus.inference.azure.com/",
    credential=DefaultAzureCredential(),
)

response = client.chat.completions.create(
    model="anthropic-claude-opus-4-7",  # the Foundry-registered name
    messages=[
        {"role": "user", "content": "Summarize the value of Claude in Microsoft Foundry in two sentences."}
    ],
    max_tokens=512,
)

print(response.choices[0].message.content)

# The API contract is similar to OpenAI's; switching models is a name change
# Same client, same call pattern, different model name in the parameter

# Costs flow through your Azure subscription
# Tracing visible in Azure Portal under your Foundry resource

For organizations migrating workloads from direct Anthropic API to Foundry-hosted Claude, the migration is largely a configuration change. The Anthropic Messages API and Foundry’s chat-completions API have similar structure; some prompt-engineering patterns translate cleanly, others may need small adjustments. Plan to validate prompts and outputs against your golden set before full cutover; treat the migration like any model-provider change with regression testing.

# IAM policy snippet for least-privilege Claude usage in Foundry
{
  "Version": "2026-05-01",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "Microsoft.AIFoundry/inferences/create",
        "Microsoft.AIFoundry/inferences/read"
      ],
      "Resource": "/subscriptions/{subId}/resourceGroups/{rg}/providers/Microsoft.AIFoundry/accounts/{account}/models/anthropic-claude-*"
    }
  ]
}

# This grants only inference rights on Anthropic models
# Production deployments should use scoped roles, not user accounts
# Use Azure managed identities for service-to-service calls

How Claude on each cloud compares in 2026

The Claude-on-cloud landscape now has three meaningful options. The table below maps the differences.

Option	API Contract	Billing	Auth	Best For
Claude in Microsoft Foundry	Foundry chat-completions API	Azure subscription	Azure AD	Azure enterprises wanting multi-model platform
Claude Platform on AWS	Anthropic Messages API (full)	AWS Marketplace	AWS IAM	AWS enterprises wanting full Anthropic features
Claude in Amazon Bedrock	Bedrock API (Anthropic subset)	AWS billing	AWS IAM	Bedrock-native teams, multi-model Bedrock workflows
Claude in Google Vertex AI	Vertex AI prediction API	GCP project	GCP IAM	GCP enterprises consolidating on Vertex
Direct Anthropic API	Anthropic Messages API (full)	Anthropic invoice	Anthropic API key	Anthropic-native teams, fastest feature access

The decision framework for picking a Claude hosting option in 2026. Use Foundry if you’re Azure-native and want Microsoft’s unified multi-model platform with Azure billing. Use AWS Claude Platform (or Bedrock) if you’re AWS-native; pick Claude Platform for full feature access, Bedrock for unified multi-model API. Use Vertex AI if you’re GCP-native. Use direct Anthropic API if you want the fastest access to new features and your procurement situation allows a separate vendor relationship. Most enterprises pick based on where their existing infrastructure lives.

Feature parity across the four cloud-hosted options matters too. New Claude features (Agent Skills, Managed Agents, Files API, code execution) historically reached the direct Anthropic API first and the cloud-hosted versions later. The gap has narrowed in 2026 but isn’t zero; for cutting-edge features, the direct API may be ahead by weeks to months. Plan accordingly.

What’s next

Three threads to track over the next 90 days. First, feature parity. Will Claude in Microsoft Foundry catch up to the direct Anthropic API on features like Managed Agents and Agent Skills? The pace at which new Claude features arrive in Foundry will signal Microsoft’s investment in the partnership.

Second, the OpenAI competitive response. With Claude now natively available in Foundry, OpenAI loses default-vendor positioning inside Microsoft’s developer ecosystem. Watch for OpenAI to introduce Foundry-specific incentives — pricing, exclusive features, deeper Microsoft integrations — to defend its position.

Third, the broader multi-provider trend. If Microsoft genuinely commits to multi-provider AI in Foundry (more models, faster feature parity, fewer single-vendor advantages), it pressures Google Cloud to do the same in Vertex AI. The competitive dynamic moves toward “best multi-model platform” rather than “best single-provider relationship.” That’s a meaningful strategic shift for the cloud-AI landscape.

For Anthropic specifically, the Foundry launch is part of a broader distribution-expansion strategy. The combination of direct API, AWS-native platform, AWS Bedrock, Microsoft Foundry, and Google Vertex AI gives Anthropic distribution reach comparable to OpenAI’s for the first time. With the company reportedly raising at $900B+ valuation, distribution depth supports the elevated valuation. The next 12 months will reveal whether Anthropic can convert distribution reach into enterprise revenue growth comparable to OpenAI’s.

Frequently Asked Questions

How is Claude in Microsoft Foundry different from Claude on AWS or GCP?

Different procurement, authentication, and billing paths. The model itself is the same; the surrounding platform integration differs. Foundry uses Azure billing and Azure AD; AWS uses AWS billing and IAM; GCP uses GCP project billing and IAM. Feature parity across these has improved but isn’t perfect; new Claude features sometimes appear in one cloud before others.

Does Microsoft now favor Claude over OpenAI’s GPT?

No. Microsoft remains heavily invested in OpenAI; that relationship continues. Adding Claude to Foundry is a multi-model strategy, not a switch. The framing is “we offer the best AI models from multiple providers” rather than “we prefer Anthropic.” OpenAI’s GPT family stays as a first-class option in Foundry.

What’s the pricing for Claude in Microsoft Foundry?

Per-token pricing largely matches Anthropic’s direct API pricing, with Azure procurement on top. For enterprises with Azure EDPs, effective pricing may be lower due to discount programs. For pay-as-you-go customers, expect prices similar to direct Anthropic API. The latest pricing is in the Azure portal’s Foundry pricing page; check there for current rates.

Can I migrate from OpenAI GPT to Claude in Foundry without rewriting my application?

Largely yes for the API surface. The Foundry chat-completions API is similar between OpenAI and Anthropic models; switching is mostly a model-name change. But prompts often need adjustment — Claude and GPT have different prompting styles, and what works well on one may need tuning for the other. Plan to validate prompts before declaring the migration complete.

Does Claude in Foundry support agent capabilities?

Basic agent patterns work via standard chat-completions API. Anthropic-specific agent features (Agent Skills, Managed Agents, code execution) may have phased availability in Foundry — check the Foundry release notes for current support. For full Anthropic agent capability, the direct Anthropic API and Claude Platform on AWS remain ahead in feature support as of mid-2026.

What about data privacy and residency?

Customer data stays within Azure’s compliance boundaries. Foundry inference happens in Azure regions you select; data residency follows your Azure subscription’s configuration. Microsoft’s enterprise contracts apply (HIPAA, SOC 2, FedRAMP, etc. as applicable). The Foundry data-handling commitments are part of your existing Azure agreement; no additional Anthropic contract required.

What does this mean for the OpenAI exclusivity narrative?

The OpenAI-Microsoft relationship remains close — OpenAI’s primary cloud is Azure, and Microsoft has substantial OpenAI equity. But the “Azure means OpenAI exclusively” narrative was always more journalistic shorthand than reality. Foundry’s multi-model strategy explicitly admits OpenAI is one option among several. The strategic shift is real but largely overdue.