The Single-Model Era Is Over

Microsoft built its original Copilot offering entirely on OpenAI's models. That made sense in 2023, when Microsoft had made a multi-billion-dollar bet on OpenAI and the two companies were deeply intertwined. But the AI model landscape has changed dramatically since then, and Microsoft's strategy has changed with it.

As of March 2026, Anthropic's Claude is available across the full Microsoft 365 Copilot experience. Users on the Frontier program can now select Claude as their model in mainline Copilot Chat, not just in the specific features — Researcher and Excel — where it was previously available. And Copilot Cowork, Microsoft's flagship new agent product, is powered by Claude as its reasoning engine.

This is a significant strategic shift worth understanding clearly.

Why Microsoft Made This Move

The honest answer is competitive pressure combined with genuine technical differentiation. Microsoft's chief marketing officer for AI at Work, Jared Spataro, put it plainly: "Every 60 days at least, there's a new king of the hill." The pace of model improvement means that betting everything on a single model provider is increasingly risky. Customers want the best model for each task, not the best model available from a single vendor.

Microsoft's response is to build what they describe as a model-diverse platform — one that allows customers to route different tasks to different models based on performance rather than vendor loyalty. The architecture that makes this possible is the same infrastructure running Copilot Studio: a model-agnostic orchestration layer that can direct queries to any supported model.

Claude's integration into this platform reflects both the deepening Microsoft-Anthropic commercial relationship — including a $30 billion Azure compute deal announced in November 2025 — and genuine capability advantages Claude brings to specific enterprise use cases.

How Model Selection Works in Practice

In Copilot Studio, developers and administrators can now select a model per agent through a dropdown. An organisation might configure a research agent to use Claude for its long-context synthesis capabilities while keeping a customer-facing chat agent on a faster OpenAI model optimised for real-time response. The routing happens at the agent configuration level, not at the user prompt level.

In mainline Copilot Chat on the Frontier program, users can switch between Claude and OpenAI models directly. Audit logs and compliance controls apply regardless of which model is selected, and all data processing stays within Microsoft's tenant boundaries because Anthropic operates as a subprocessor within that arrangement.

For enterprise deployments in the EU and UK, availability currently requires an opt-in step due to regional data governance requirements. Most commercial tenants gained default access in early January 2026.

When Claude Is the Right Choice

Not all tasks benefit equally from Claude. Understanding where Claude's capabilities translate into better outcomes helps organisations configure their agents intelligently rather than just picking a model arbitrarily.

Claude's strongest advantages in the M365 context are long-context reasoning and complex document synthesis. Claude Opus 4.6 supports a one-million-token context window and up to 128,000 tokens of output — specifications that matter when an agent needs to read and synthesise large volumes of documents, legal agreements, financial reports, or research. For agents handling tasks like cross-document analysis, proposal generation from multiple source materials, or due diligence workflows, Claude's performance in these areas is meaningfully stronger.

Claude also performs well on instruction-following in agentic workflows — the ability to reliably execute a multi-step plan without drifting from the original intent. For Cowork agents handling consequential business processes, this matters more than raw benchmark scores.

OpenAI's models remain strong for real-time conversational tasks, code generation, and structured data extraction where response speed is a priority. The optimal enterprise configuration is likely model selection by use case rather than a single-model policy.

What This Means for Organisations Building Agents Today

If your organisation is currently building or evaluating custom agents in Copilot Studio, the arrival of Claude as a first-class model option changes the evaluation criteria. The question is no longer "how do we get the most out of GPT in our workflows?" It is "which model performs best for each specific task in our agent portfolio?"

That is a more sophisticated question, but it is also a more productive one. The organisations that build model-selection discipline into their agent strategy now will have a significant advantage as the model landscape continues to evolve.

At Trim Journey, we help organisations navigate exactly this kind of agent architecture decision — including when to use Claude, when to use OpenAI, and how to build Copilot Studio agents that route intelligently based on task type. Book a 30-minute call to talk through your agent strategy.