
Microsoft has started rolling out OpenAI's new GPT-5.6 model across Microsoft 365 Copilot. In plain terms, the AI behind Copilot in Word, Excel, PowerPoint, Chat and Cowork just got a significant upgrade, and Microsoft is calling it the preferred model for Copilot.
Most firms won't need to do anything to get it. But there's a decision buried in this announcement that's worth understanding, especially if your firm handles sensitive client data.
Copilot doesn't run on one AI model. Behind the scenes, Microsoft now offers a choice between models from OpenAI, the company behind ChatGPT, and Anthropic, the company behind Claude.
In our training work we've found that most teams get better results from Claude on the heavier tasks, things like complex Excel analysis or Cowork, the newer part of Copilot where you describe what you need and it goes away and produces the finished piece of work.
Using Claude in Copilot requires your IT admin to switch it on, and that switch comes with a data question. Claude runs on Anthropic's systems rather than Microsoft's. It's covered by Microsoft's data protection terms, but your Copilot data is processed outside Microsoft's own environment, and for UK firms it's switched off by default.
Plenty of our clients have looked at this properly and switched it on. Others have decided their client commitments or regulatory position means everything stays inside Microsoft, full stop. Both are sensible positions. The frustration for the second group has been missing out on the best models for the most useful parts of Copilot. We covered the governance side in more detail in our Cowork, Agent 365 and E7 explainer.
This is the first OpenAI model that's been properly built for that same heavy, multi-step work, and it runs entirely inside Microsoft's environment. No settings to change, no data decision to make.
So if your firm said no to Claude, you're no longer locked out of Copilot's most capable features. And if you said yes, you now have two strong options and Copilot will often pick the right one for the task automatically. Where the model selector is available, you can also choose GPT-5.6 directly.
Worth knowing too: Microsoft is building its own AI models, and some are already appearing quietly inside PowerPoint and Teams. The direction of travel is more choice, with more of it staying inside Microsoft's walls.
Pick a couple of real tasks your team does regularly, a monthly reporting pack, a client letter, tidying up a messy spreadsheet, and try them in Copilot this week. If you have access to both models, try both. The differences are bigger than people expect, and which model wins depends on the task, not the branding.
And if nobody in your firm can tell you whether Claude is switched on or off, that's worth finding out. Whichever way you've gone, it should be a decision someone made deliberately, not a default nobody noticed.
Availability is rolling out gradually and may vary by tenant, so don't worry if you can't see GPT-5.6 in the selector yet.
At IQIT we help UK firms get real value from Copilot, from hands-on training to working out what your model and data setup should look like. If you'd like a straight answer on where your firm stands, get in touch.
Book a free consultation to talk through where you are and where you want to be. No pressure, no hard sell. Just an honest conversation.