OpenAI Calls GPT-5.6 Microsoft's "Preferred Model" — Right After Reports of a Cost-Cutting Split
The announcement lands days after Bloomberg reported Microsoft was quietly swapping in its own AI to power Word and Excel.
OpenAI picked an interesting week to remind everyone how close it is with Microsoft.
On Thursday, during the launch of its new GPT 5.6 model, OpenAI announced that the model would serve as the "preferred model" powering Microsoft 365 Copilot, the AI assistant baked into Word, Excel, PowerPoint, and Cowork. In a blog post announcing the news, OpenAI framed it as a natural extension of its long-running relationship with Microsoft, writing: "Our partnership with Microsoft has always been about bringing the benefits of advanced AI to more individuals and organizations, and we're excited to continue building on that shared commitment."
That sounds like standard corporate-partnership language, until you consider the timing. Just days earlier, Bloomberg reported that Microsoft had started swapping out some of OpenAI's technology for its own in-house models, known as MAI, inside apps like Word and Excel. The stated goal, according to that reporting, was cutting costs.
That story reignited a question that's been simmering for a while now: is the OpenAI-Microsoft relationship, once treated as one of the tightest and most consequential alliances in tech, starting to come apart at the seams? The two companies have gone from looking practically fused at the hip to sending what can charitably be described as mixed signals about where things stand.
A Preferred Model, But Preferred For What, Exactly
OpenAI's GPT 5.6 announcement reads like an attempt to squash any speculation about a breakup before it gathers more momentum. But the actual substance of the "preferred model" designation is thin. Beyond the fact that OpenAI's technology will keep running inside Microsoft's productivity apps, it's not clear what the label changes in practice, how it's defined contractually, or what obligations, if any, it creates for Microsoft going forward.
That vagueness matters, because the Bloomberg report and OpenAI's new announcement aren't actually in direct conflict. Nobody reported that ChatGPT-based tools were being ripped out of Microsoft's products entirely. The reporting was more specific than that: Microsoft was leaning more heavily on its own MAI models to handle certain tasks, presumably because running its own software is cheaper than paying for OpenAI's. OpenAI declaring itself the "preferred" option for Copilot doesn't really contradict that. A company can still be the preferred vendor while a partner increasingly builds and uses its own alternative for cost reasons.
Two Companies, One Complicated Relationship
The backdrop here is a partnership that has always been unusual in scale and structure. Microsoft has poured enormous resources into OpenAI and built much of its AI product strategy, from Copilot to its cloud offerings, around access to OpenAI's models. For a long stretch, the two were talked about almost as a single entity in the AI conversation.
But that dynamic has been shifting. Microsoft has been investing heavily in its own MAI models, which gives it leverage it didn't have before: the ability to route certain workloads to its own technology instead of paying OpenAI for access. That's a meaningful change in posture for a company that once looked entirely dependent on its partner for frontier AI capability.
None of this means the partnership is ending. OpenAI's software still runs inside Microsoft's flagship productivity suite, and that's not a small thing to walk away from for either side. But the fact that OpenAI felt the need to make a public statement asserting its continued relevance to Copilot, timed right after reporting suggested Microsoft was pulling back, tells you something about how sensitive both companies are to the narrative right now.
What to Watch
The real signal to track isn't the marketing language around GPT 5.6. It's how much of Copilot's actual workload ends up running on MAI versus OpenAI's models over the coming months, and whether Microsoft keeps expanding its in-house capability at OpenAI's expense. "Preferred model" is a nice title. Whether it reflects growing distance or genuine continuity is something only Microsoft's product decisions, not OpenAI's press releases, will actually answer.
