GPT 5.6 now preferred model in Microsoft 365 Copilot

OpenAI has confirmed that GPT-5.6 is now the preferred model powering Microsoft 365 Copilot, running across Word, Excel, PowerPoint, Outlook, Teams Chat and the newer Cowork feature set. The announcement, published on OpenAI's site, frames the update as a quality and speed upgrade for the millions of users who rely on Copilot inside Microsoft's productivity suite — including a growing number of Australian mid-market businesses that have adopted Microsoft 365 as their default operating environment.

For a market where "Copilot for business Australia" has become one of the more searched AI terms this year, the practical question isn't whether GPT-5.6 is smarter in the abstract. It's what changes for someone drafting a board pack in Word, building a pricing model in Excel, or asking Copilot Chat to summarise a supplier contract on a Tuesday afternoon.

What OpenAI is actually claiming about GPT-5.6 inside Copilot

According to OpenAI's announcement, the model swap improves reasoning, response speed and output quality across the core Microsoft 365 apps. Copilot has run on various OpenAI models since Microsoft first embedded generative AI into Word and Excel; GPT-5.6 is presented as the next step in that lineage, not a separate product. Users don't install anything or choose a model from a dropdown — Microsoft handles model selection on the back end, which means the upgrade lands invisibly for most Copilot licence holders.

That invisibility is worth pausing on. Microsoft has built Copilot's commercial pitch around abstraction — buy the licence, get the assistant, don't worry about which model sits underneath. OpenAI's decision to publicise the model change publicly, on its own site, is a departure from that quiet default. It's OpenAI reasserting its brand inside a product most users experience as "Microsoft's AI", at a time when Microsoft has been diversifying its own model sourcing and building smaller in-house models through Microsoft AI.

Word, Excel and PowerPoint are where the difference will actually show up

The apps named in the announcement — Word, Excel, PowerPoint and Chat — are the ones Australian office workers touch daily, which is exactly why the update matters more than a typical model release. A stronger model in Excel means formula generation and data analysis prompts should hold up better against messy, real-world spreadsheets rather than clean demo data. In PowerPoint, it should mean fewer generic slide drafts that need a full rewrite. In Word, it plays into document drafting and summarisation — the two tasks that show up most often in how mid-market firms actually use Copilot.

Cowork, the newer collaborative feature Microsoft has been rolling into Copilot, is also named specifically. It points to Microsoft's broader push to have Copilot operate across multiple documents and threads at once, rather than responding to single prompts in isolation — a shift toward the kind of multi-step, judgement-heavy work that has so far been harder for AI assistants to handle reliably.

Why the model underneath actually matters to a business paying for the licence

Most Australian businesses evaluating Microsoft 365 Copilot treat it as a single decision — turn it on or don't — without much visibility into which model is doing the work at any given time. That's understandable; Microsoft doesn't make it a selling point. But it has a real consequence for procurement and platform strategy: a business that assessed Copilot's output quality six months ago and found it underwhelming for a specific task, such as drafting client correspondence or reconciling monthly reports, is now testing against a materially different model without having changed anything on its end.

This is the kind of detail that matters more in artificial intelligence consulting conversations than it does in a Microsoft sales deck. Platform selection isn't a one-off decision — Copilot, Claude and ChatGPT for Work all improve on different timelines, and a workflow that was rejected as "not good enough for AI" a year ago may now be worth revisiting on the same platform, simply because the model changed underneath it. That's a genuine argument for treating AI adoption as an ongoing review rather than a single deployment decision, particularly for businesses that already pay for Microsoft 365 E3 or E5 and are deciding whether to add Copilot licences on top.

The bigger tension is who actually controls the assistant

There's a structural question sitting underneath this announcement that goes beyond one model upgrade. Microsoft has invested heavily in OpenAI, but it has also been building its own foundation models and has previously flagged interest in reducing single-vendor dependency inside Copilot. OpenAI publicising itself as the "preferred model" reads as a statement about the current state of that relationship — a reminder, aimed partly at enterprise customers and partly at Microsoft itself, of whose technology is doing the heavy lifting inside one of the world's most widely deployed AI products.

For Australian businesses, that dependency runs one level deeper than most licence holders realise. A Copilot subscription bought through a Microsoft partner in Sydney or Melbourne is, in practice, a bet on OpenAI's model roadmap as much as it is a bet on Microsoft's product roadmap — and that arrangement can shift again, as it apparently already has. Whether GPT-5.6 stays the preferred model for the next update cycle, or gets swapped out for a Microsoft-built alternative, is not something the businesses paying for Copilot licences currently get much say in. That's the part of this story still worth watching.