OpenAI Names GPT-5.6 the ‘Preferred Model’ for Microsoft 365 Copilot as Split Rumors Swirl

OpenAI said on Thursday that its newly released GPT-5.6 will serve as the "preferred model" powering Microsoft's 365 Copilot, a public reassurance issued just days after reports suggested the two companies were loosening their once-tight bond.

The announcement came during the launch of GPT-5.6 and was accompanied by a blog post confirming the model would support Microsoft users across productivity apps including Word, Excel, PowerPoint, and Cowork. "Our partnership with Microsoft has always been about bringing the benefits of advanced AI to more individuals and organizations," OpenAI wrote in the post.

The timing is the story. Earlier in the week, Bloomberg reported that Microsoft was replacing some of OpenAI's software with its own in-house models, known as MAI, to reduce costs, with those models being deployed to power apps such as Word and Excel. That report fed a running narrative about whether the two firms, long treated as inseparable, were quietly drifting apart.

What 'preferred model' actually means

Less than the framing implies. OpenAI's language confirms that its software will keep running inside Microsoft's apps, but the term "preferred model" is not defined beyond that. It does not describe exclusivity, revenue terms, or the scope of how many workloads GPT-5.6 will handle versus alternatives.

Crucially, the new disclosure does not contradict the earlier reporting. Bloomberg's account never claimed that OpenAI's software would stop powering Microsoft's apps. It said Microsoft was leaning more heavily on its own MAI models to trim expenses. A model can be "preferred" in some contexts while an in-house system quietly picks up others. The two statements can both be true, which is why the announcement reads more as reputation management than a change in strategy.

The commercial logic behind the tension

For Microsoft, the incentive to build and route traffic through its own models is straightforward. Serving frontier-model inference at the scale of Word, Excel, and Copilot is expensive, and every workload handled by an in-house system is a workload Microsoft does not pay a partner to run. That cost calculus does not require a falling-out to explain. It is the natural behavior of a company that has invested billions into a partner and now wants leverage over its own margins.

For OpenAI, Microsoft remains one of its most important distribution channels and financial backers. A public signal that GPT-5.6 stays central to Copilot protects a relationship that underpins a meaningful share of OpenAI's enterprise reach.

The announcement also arrived on a busy day for OpenAI. The company launched its GPT-5.6 family of models and, separately, confirmed that Fidji Simo was stepping down from her role as the company's number two, according to TechCrunch's reporting.

Why it matters

Enterprises standardizing on Copilot have a practical interest in knowing whose models sit underneath their tools, since that affects performance, data handling, and pricing over time. A "preferred model" label offers reassurance without commitment. Until Microsoft and OpenAI disclose the split of workloads between GPT-5.6 and MAI, buyers are left reading the same tea leaves the rest of the market is: a partnership that both sides insist is intact, described in language vague enough to accommodate a slow divergence.