Claude in Microsoft 365 Copilot: what it means for your SMB

That is genuinely interesting. It is also smaller than the headline makes it sound. Here is the operator read on what it changes for an SMB on Microsoft 365, and what it does not.

What actually changed

Three things shipped in quick succession, and it helps to keep them separate.

• Model choice in Copilot. Claude Opus 4.8 is now in the model selector across Copilot Chat, Excel, PowerPoint, Copilot Studio and the Cowork experience, for eligible licensed users. Which surfaces you see can depend on your tenant and release ring.
• Agentic Office apps. Microsoft brought Word, Excel and PowerPoint agents to general availability on 22 April 2026 - agents that do multi-step work inside the documents themselves, not just chat beside them.
• Microsoft Agent 365. Generally available from 1 May 2026 at about US$15 (roughly AUD $23) per user per month, this is the layer for managing agents at scale across an organisation.

The piece that matters most for the model question is the first one. You can now pick Anthropic’s brain for a given task inside Microsoft’s product. You do not have to leave Copilot, and you do not pay extra for the privilege of choosing Claude over the default.

Claude vs the default OpenAI model inside Copilot

The instinct is to ask which model is better. The more useful question is which is better for what, because the gap is narrower than the marketing on either side suggests.

Anthropic pitches Opus 4.8 at complex, multi-step coding, long-horizon agentic work, document drafting, data analysis and presentations, with close instruction-following and a tendency to flag its own uncertainty rather than bluff. In practice, a lot of teams find Claude’s writing tone and its carefulness suit business documents - proposals, reports, client emails - well. Microsoft’s default OpenAI models are fast, capable and perfectly good for the bulk of everyday Copilot work.

For routine use - summarising a thread, drafting a quick email, asking a question of a document - you will struggle to notice the difference. Where it shows up is at the heavy end: a long or structured document, dense analysis in Excel, a multi-step agent in Copilot Studio, anything where following the instruction precisely and getting the tone right actually matters.

The right way to decide is not a benchmark. It is to run both on your real work for a week and let the people doing the work pick. Whoever writes your proposals will know within a few documents which one they trust more. That beats any leaderboard.

When it is worth switching the model

Treat the model selector as a tuning knob for your highest-value tasks, not a setting you sweat over on every prompt.

Switch to Claude when the output quality is worth tuning: long or complex writing, careful data work, agentic tasks, or anything client-facing where the tone has to land. Leave the default in place for quick chat, summaries and routine drafting, where it is fine and the switch is not worth the friction. The mistake is making this a per-prompt ritual. Pick the model for the kind of work, set it, and get on with it.

The cost picture

Here is where SMB owners need to be clear-eyed. Choosing Claude inside Copilot is free - it is included in the licence. The cost that matters is Copilot itself.

Microsoft 365 Copilot runs about US$30, roughly AUD $46, per user per month, and that sits on top of your existing Microsoft 365 subscription. For a ten-person team that is around AUD $5,500 a year before you have touched the underlying Microsoft 365 bill. Add Microsoft Agent 365 at about AUD $23 per user per month if you go down the managed-agents path, and the numbers grow again.

None of that changes because Claude is now an option. The model choice is a no-cost upgrade to a seat you are already paying for. The investment decision was always the Copilot seat, and it still is.

The honest take

The model matters far less than whether your team has actually adopted Copilot at all.

This is the part the announcements skip. Across Australian SMBs, the common story is not “we picked the wrong model”, it is “we bought Copilot, a few people tried it, and now it sits there”. A better model underneath an unused tool changes nothing. Adoption is the whole game: pick a handful of high-value workflows, train people properly on them, and put someone in charge of the rollout so it does not quietly die.

Once Copilot is genuinely in the daily flow of work, switching your heavy tasks to Claude is a sensible bit of tuning that can lift the quality of your most important documents. Before that, it is a distraction dressed up as a decision.

So if you are on Microsoft 365 and weighing this up: do not let the Claude headline pull your attention. Fix adoption first. Then, on the work that warrants it, reach for the model that does it best - which, increasingly, you get to choose.