Why Microsoft is betting on model choice instead of picking one AI winner

Picture this: your manager asks for a status update on a project, and the real answer is scattered across 50 Word docs, 17 Teams channels, and a few dozen emails nobody bothered to CC you on. No single AI model, no matter how smart, can answer that question if it can't see where the information actually lives. That's the problem Microsoft is trying to solve with the new Copilot — and it's a completely different bet than the one OpenAI, Anthropic, and Google are making with their own AI work tools.

While ChatGPT, Claude Code, and Gemini each ship as a tightly bundled app-plus-model experience, Microsoft is doing something structurally different: it's decoupling the model from the product and adding a third layer most people have never heard of, called Work IQ. Here's how it actually works, and how you can start using it today.

What makes Microsoft's Copilot strategy different?

Every major AI lab builds what's sometimes called a "harness" — the application layer that sits on top of a language model. ChatGPT runs on OpenAI's models, Claude Code runs on Anthropic's models, Gemini runs on Google's models. For years, Copilot worked the same way, running almost exclusively on OpenAI.

That's changed. Microsoft has turned Copilot into a multi-vendor platform. Microsoft Copilot provides access to multiple frontier models, giving users the flexibility to choose the one that best fits the work they need to accomplish — but model choice is only part of what drives a great response, because equally important is what the model knows about your work. That second part is Work IQ, and it's arguably the more important of the two.

As one analysis of the rollout put it, Microsoft's Copilot has matured from a single-engine novelty into a platform for intelligent routing and specialization — a change that will define how enterprises deploy AI going forward.

How does model choice actually work inside Copilot?

The headline feature is simple on the surface: you're no longer locked into one AI provider. Microsoft has announced a shift to a multi-model architecture that brings Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Sol directly into the Copilot experience across Word, Excel, PowerPoint, Chat, Cowork, and Copilot Studio.

You can try this yourself right now inside Microsoft 365 Copilot. The model picker shows up differently depending on which app you're in:

In Excel: before submitting a prompt, you can select which model you prefer to use — Copilot in Excel supports the latest Anthropic and OpenAI models, and you switch between models using the model picker menu in the upper right corner of the Copilot pane. By default, the model picker is set to Auto and the model is automatically chosen for you.

In PowerPoint: open PowerPoint, select Home > Copilot to open the Copilot pane, make sure you're in editing mode, then select the AI model switcher to see available options — including Auto, ChatGPT, or Claude. Worth noting: your model choice is only for the active session, so when you close PowerPoint or end your session, Copilot reverts to Auto.

This "Auto" option matters more than it sounds. Rather than making every employee become an amateur model researcher, Microsoft's reasoning is that no single model is ideal for every task, so the company embraces a flexible, multi-model approach that lets the user or Copilot choose what is the best or most appropriate model for the job. Different models get described as having different strengths — for example, Claude Opus brings advanced reasoning and long-context understanding, while GPT models deliver high-speed creativity, synthesis, and generative fluency.

This wasn't a fast pivot, either. Microsoft has been building toward this for over a year. Microsoft announced on September 24, 2025 that Anthropic's Claude Sonnet 4 and Claude Opus 4.1 are now available as selectable model options inside Microsoft 365 Copilot, and the model lineup has kept expanding since — new Claude and GPT versions have rolled in roughly every few weeks.

What is Work IQ and why does it matter more than the model itself?

Here's the part that's easy to miss if you're only paying attention to which chatbot brand is "winning." Microsoft calls it Work IQ, and it's not a product you open — it's an invisible layer sitting underneath everything Copilot does.

Work IQ is not a new application or service that users interact with directly — it's a shared intelligence layer that continuously interprets work happening across the tenant, and that distinction is critical because it explains why Work IQ shows up everywhere Microsoft 365 Copilot works.

Microsoft's own support documentation breaks it down for end users: your work data — the emails, files, chats, meetings, people, projects, accounts, and opportunities you already work with across Microsoft 365 apps — is powered by Work IQ. Instead of forcing you to manually choose between searching your files or searching the web, the separate Work and Web tabs in Copilot have been replaced by the Work IQ button, so instead of switching between work grounding and web grounding, you now use a single on/off control to decide whether Copilot can reference your work data.

Under the hood, Microsoft describes Work IQ as having three layers working together: Data, Memory, and Inference, which work together to give Microsoft Copilot and your agents continuous, contextual understanding of work — unifying signals from files, emails, meetings, chats, and business systems across Microsoft 365 to capture how work happens across the organization.

That's the technical answer to the "50 Word docs, 17 Teams channels, dozens of emails" problem. Instead of you manually searching through every surface where a conversation might have happened, Work IQ is designed to surface the pieces that are actually relevant — and only the pieces you're allowed to see.

Does Work IQ respect existing permissions and security?

This is the question every IT department asks first, and it's a fair one. Microsoft's documentation is explicit that Work IQ doesn't create new access — it just makes existing access more usable. Copilot respects your existing Microsoft 365 permissions — when Work IQ is on, Copilot can only reference work content you're already authorized to access, and it applies your organization's security and governance controls to every request.

On the data-retention side: your data stays yours, meaning Copilot doesn't train public models on your company's data, and it doesn't retain your prompts to train models outside your organization. And critically, your permissions are always enforced — Work IQ works with your organization's data in place and honors the same access rules you already have, without creating separate, over-shared copies of your content.

If you're a developer rather than a day-to-day Copilot user, there's also a Work IQ CLI that exposes this same context layer to coding tools. Microsoft Work IQ is a command-line interface and Model Context Protocol server that connects AI assistants to your Microsoft 365 Copilot data, enabling you to query emails, meetings, documents, Teams messages, and more using natural language. That means a coding assistant working on a feature can pull in the actual meeting notes or email thread where that feature was discussed — the MCP server mode allows your AI assistant to automatically pull in relevant workplace context when you're working on code, so if you're implementing a feature discussed in a recent meeting, your AI assistant can access that meeting context to provide more relevant suggestions.

How is this different from how ChatGPT, Claude, and Gemini are built?

The honest answer: those tools were built single-player first. ChatGPT, Claude Code, and Gemini all started as experiences for one person working alone, and enterprise features got bolted on afterward. Copilot's bet is the opposite — build for the organization from day one, then make it feel good for individuals.

That shows up in features like Cowork, Microsoft's agentic work product. Copilot Cowork is designed to deliver lower cost from a runtime that efficiently finds the right information and tools, model choice that matches the right model to each task, and billing that charges you only for what you use — and you are not locked into one model, you can use the most efficient model or frontier models. The full package bundles everything together: the Microsoft 365 Copilot license includes Copilot Chat, Copilot in Word, Excel, PowerPoint, Outlook, and Teams, the Work IQ context engine, a multi-model system offering frontier intelligence, pre-built agents like Researcher and Analyst, and custom agents built with Agent Builder.

One analyst comparison even found a cost edge: when comparing the cost per prompt between Copilot Cowork and Claude Cowork with their Microsoft 365 connector, testing showed that Copilot Cowork on average was 30-40% cheaper.

How do you actually try this out today?

If your organization already has a Microsoft 365 Copilot license, you don't need to wait for anything special — the features described above are either live or rolling out now. A few practical starting points:

1. Open Microsoft 365 Copilot Chat and look for the model picker — it's typically in the top corner of the chat pane.
2. In Excel or PowerPoint, open the Copilot pane and check for the AI model switcher described above. If you don't see it, it may have moved under Settings.
3. Toggle the Work IQ button on or off to control whether Copilot references your organizational data or sticks to general knowledge.
4. If you're technical, explore the Work IQ CLI documentation to connect your coding assistant directly to your company's Microsoft 365 context.
5. Keep an eye on the AI at Work Roadmap — Microsoft ships new model additions and Work IQ capabilities on close to a weekly cadence right now.

The strategy here isn't really about picking a favorite AI model. It's a bet that most companies don't actually care which lab built the model answering their question — they care whether that answer is right, grounded in what's actually happening at their company, and safe to share. Whether that bet pays off depends less on which model wins the next benchmark and more on whether Work IQ can keep making sense of the sprawling, messy reality of how people actually work together.