Claude arrives in Microsoft Foundry and Microsoft 365 Copilot

Claude is now available in public preview within Microsoft Foundry and Microsoft 365 Copilot. The integration lets companies use Anthropic’s models through their existing Azure tools, without separately contracting and managing another AI system.

What companies can do

Azure customers can access Claude Sonnet 4.5, Claude Haiku 4.5 and Claude Opus 4.1 through Microsoft Foundry, Microsoft’s platform for building enterprise applications and agents. Anthropic manages the infrastructure through a serverless deployment, while developers build their solutions using Foundry’s APIs and tools.

In practice, a company can create an agent that reviews documents, automates customer service tasks or helps with programming without leaving its Azure environment. The models can also be used with Claude Code, Anthropic’s agent for software development.

The integration works with existing Azure agreements and is eligible for Microsoft Azure consumption commitments, known as MACC. That can eliminate the need for new vendor approvals, contracts and separate billing systems, a process that can delay adoption by weeks or months in large companies.

Claude is initially available in the Global Standard region, with Anthropic’s regular API pricing. The US DataZone option, which limits processing to a US data zone, will arrive later.

Claude also comes to Excel

The update is not limited to Foundry. Microsoft 365 Copilot already uses Claude for its Researcher agent, designed to conduct complex, multi-step research, and lets you create custom agents from Copilot Studio.

In addition, Excel’s Agent mode now includes a preview option for working with Claude directly in spreadsheets. It can:

• Generate formulas.
• Analyze data.
• Detect errors.
• Test and adjust solutions within the file.

For example, you could ask it to find inconsistencies in a sales table, create a formula to calculate margins and modify the report without copying the data into another application.

Which model fits each task

Microsoft offers all three models with features such as code execution, web search, citations, vision, tool use and temporary instruction storage to reduce costs and speed up responses.

• Sonnet 4.5 is designed for programming, complex reasoning and agents that can complete multi-step tasks.
• Haiku 4.5 prioritizes speed and cost. Anthropic says it delivers performance close to that of the most advanced models at one third of Sonnet’s price, making it suitable for automated support, moderation and high-volume applications.
• Opus 4.1 is intended for specialized problems that require extended analysis and attention to detail.

For you, the most important change is not simply that Claude is appearing in more products. Companies that already pay for Azure or Microsoft 365 can try these capabilities without rebuilding their entire procurement, security and billing structure. Competition among AI providers is moving into a very specific area: who can integrate most effectively with the tools companies already use every day.