Claude Opus 5 arrives in Microsoft 365 Copilot

On July 24th, 2026, Microsoft released another update: welcome to Opus 5.

They’ve added the latest flagship Anthropic model, Claude Opus 5, to the model selector. It’s rolling out across Word, Excel, PowerPoint, Copilot Chat, Copilot Cowork, and Copilot Studio, subject to region and tenant configuration. So don’t panic if you haven’t noticed anything different; it’s just down to those two factors.

At Changing Social, we focus on the adoption and transformation the Microsoft ecosystem drives through people, so let’s look through a different lens. Not ‘what can the model do?’ but ‘what will actually change for the people who have to use it, and how do we get them there?’

Here’s the uncomfortable truth we see in organisation after organisation: the capability is the easy part. It’s already in the box. The value only shows up when behaviour changes – and behaviour is the part no model release solves for you.

 

What actually changed

Opus 5 is the newest model in Anthropic’s Opus family.

It improves on its predecessor, Opus 4.8, across agentic coding, professional knowledge work and long-horizon reasoning – the ability to hold a complex multi-step task together from start to finish without losing the thread. Two things make it matter for the workplace.

Firstly, Anthropic positions Opus 5 as a near-frontier intelligence at the same price as the model it replaces. Your people will get a meaningfully more capable model without a step-change in cost.

Secondly, it’s notably better at checking its own work! It verifies and iterates before it hands something back, rather than producing a confident first draft and stopping. So, no more extra prompt saying, ‘are you sure that’s correct?’ For real work, that reliability is what earns a user’s trust. And trust is what turns a one-off experiment into a daily habit.

Both of those are capability gains. Neither of them, on their own, changes how a single person in your organisation works on a Monday morning, so you can give them that reassurance.

 

The gap Opus 5 is built to close and the gap it can’t

Inside the apps your teams already live in, Opus 5 lets Copilot take on more of the steps between a rough idea and finished work, while leaving the final decision with your people:

• Word: moving from a rough idea to a complete draft, holding structure and flow together across longer documents.
• Excel: working through complex analysis with higher-quality modelling and more polished workbooks.
• PowerPoint: building and refining a deck from first outline to finished article, with more visual polish and better use of the slide.
• Copilot Chat: tackling open-ended, ambiguous problems: comparing options, structuring plans, reasoning things through.
• Copilot Cowork: staying with long-running tasks, orchestrating across tools and files, grounded in your organisation’s own context through Work IQ.
• Copilot Studio: building agents on Opus 5 for advanced reasoning, workflow automation, and multi-step tool use.

 

Already using Copilot Chat? Here’s what’s going to change for you

Most of your people already type questions into Copilot Chat every day.

So let’s be clear about what is actually different now without the fancy jargon…

You will see a new option in the dropdown. When you open Copilot Chat, you will now be able to pick Claude Opus 5 from the model menu.

Nothing you already do breaks. Your usual questions, prompts and shortcuts should all still work.

You’ll notice it most on the harder questions. For quick and simple asks, you won’t see much of a difference; this is because Opus 5 is not for simplistic tasks. It shines in the messy, open-ended stuff.

It’s better at catching its own mistakes. The new model will double-check itself before it answers the question. So, you should get fewer confident but wrong replies and more that you can actually rely on (but please always check any work that has been created using AI, and a skill that’s more needed than ever now).

It handles bigger, longer tasks without getting annoyed at you. Ask it something with lots of moving parts, and it’s better at keeping everything straight from start to finish.

It still knows your work. Just like before, it can draw on your emails, files, chats and meetings.

It costs nothing extra. Same license, same price. You’re simply getting a more capable option.

So, what does this really mean, day to day? When you hit a question that would normally make you stop and think, the kind you’d usually puzzle over yourself or hand to a colleague, that’s your cue to switch to Opus 5 and let it do the heavy lifting with you. The one thing to remember: it’s a choice, not automatic. You must pick it from the menu. Once your people know that, they’ve got everything they need.

Why this is a change management problem, not a rollout task

Two years ago, an AI capability upgrade might have felt like a project with a start and an end. That model is gone.

Opus 5 arrived roughly two months after Opus 4.8. Microsoft is turning Copilot into a model-agnostic platform where new frontier models land in the selector continuously, inside tools people already own. That’s a huge advantage; you inherit the frontier without a migration. But it quietly rewrites the adoption challenge.

You’re no longer managing a change. You’re building an organisation’s capacity to keep changing, comfortably, repeatedly, without change fatigue. That means the goal isn’t “train everyone on Opus 5.” It’s cultivating durable AI-ready habits and confidence, so the next model, and the one after that, land on receptive ground.

Concretely, that shifts the work in three ways:

From features to fluency. People don’t need to know what “long-horizon reasoning” means. They need to know which everyday tasks are now worth handing over, and how to judge the output. Model choice, when to reach for a heavier reasoning model versus a faster one, is becoming a genuine workplace literacy, the new equivalent of knowing which app to open.

From launch events to sustained enablement. A silent rollout is the most common way we see value evaporate: the capability appears, nobody changes how they work, and the licence spend quietly underperforms. Continuous model improvement needs continuous, lightweight enablement, champions, nudges, and visible before-and-after wins, not a single kick-off.

From adoption metrics to outcome metrics. “How many people tried it” tells you almost nothing. The number that justifies the investment is whether the distance between a rough idea and a shippable draft actually got shorter for the tasks that matter.

 

Trust and governance are adoption enablers, not blockers

There’s a reason we always fold governance into the adoption conversation rather than treating it as separate. People don’t adopt what they don’t trust.

Opus 5 in Copilot sits inside Microsoft’s enterprise-grade security, privacy, and compliance envelope, and through Work IQ its answers are grounded in your organisation’s own files, meetings, chats, and business data, not the open web alone. Access is something your admins opt into and govern, not something that simply appears for end users.

For a change practitioner, that’s gold. When you can tell an anxious team exactly where their data goes and who controls it, you remove the single biggest brake on individual adoption. Capability opens the door; trust is what gets people to walk through it.

 

What good looks like from here

If your organisation has Opus 5 landing in the selector, our advice is the same advice we’d give for the model after it:

1. Turn it on deliberately, and make the change visible. Pick a few high-value, multi-step tasks, a recurring report, a proposal, a board deck, and show the before-and-after. Momentum comes from proof, not announcements.
2. Build fluency and confidence, not feature knowledge. Equip people to judge when to lean on the model and how to sense-check what comes back. Confidence is what converts a curious try into a daily habit.
3. Design for continuous change, not a one-off launch. Stand up the champions, feedback loops, and enablement rhythm that let the next model land effortlessly. That capacity is the real asset.
4. Measure outcomes, and connect them back to the work. Track the time from idea to finished work; that’s the story that earns continued investment and executive sponsorship.

The frontier will keep moving on its own schedule. The organisations that pull ahead won’t be the ones with the newest model in the picker; they’ll be the ones who’ve built the habits, trust, and adoption capacity to put each new model to work the day it arrives. That capacity doesn’t build itself, and it doesn’t come in the box with the model.

That’s the work we do.