What does it mean that GPT-5.6 is the 'preferred model' for Microsoft 365 Copilot?

OpenAI designated GPT-5.6 as the primary model powering Microsoft 365 Copilot, but the term 'preferred' is vague and doesn't specify exclusivity or what percentage of Copilot inference uses it versus Microsoft's own MAI models.

Is Microsoft replacing OpenAI models with its own?

Bloomberg reported that Microsoft has been using its in-house MAI models to power some apps like Word and Excel to reduce costs, though OpenAI's GPT-5.6 is still designated as the preferred model for the flagship 365 Copilot product.

MAI is Microsoft's in-house family of AI models, developed as an alternative to OpenAI's models for certain internal applications, primarily to reduce inference costs at scale.

Does the GPT-5.6 announcement contradict the Microsoft cost-cutting reports?

Not really. Microsoft using MAI models in some apps while designating GPT-5.6 as the preferred Copilot model are not mutually exclusive — both scenarios can coexist.

Home›Large Language Models›Article

Large Language Models

GPT-5.6 Named "Preferred Model" for Microsoft 365 Copilot — But What Does That Actually Mean?

OpenAI has designated GPT-5.6 as the "preferred model" for Microsoft 365 Copilot — but the vague terminology raises more questions than it answers, especially in light of recent reporting that Microsoft is increasingly substituting its own MAI models to cut costs.

Dispatch desk·July 10, 2026·3 min read

TL;DR

OpenAI named GPT-5.6 the "preferred model" for Microsoft 365 Copilot, but the announcement appears designed to counter breakup speculation rather than negate reports of Microsoft expanding its own in-house MAI models.

What changed

OpenAI announced that its newly launched GPT-5.6 model will serve as the "preferred model" powering Microsoft 365 Copilot across Word, Excel, PowerPoint, and Cowork.

Why it matters

For enterprise builders and Microsoft 365 integrators, the ambiguity of "preferred model" leaves critical questions unanswered about which inference stack actually powers Copilot — and whether Microsoft's in-house MAI expansion is quietly shrinking OpenAI's footprint.

Editorial record

Drafted by autopilot. Reviewed when marked.

Let's get one thing straight before we dissect the PR here: OpenAI calling GPT-5.6 the "preferred model" for Microsoft 365 Copilot is not a counter-announcement. It's a vibes correction. And it's a pretty telling one.

Here's the backstory. Earlier this week, Bloomberg reported that Microsoft has been quietly swapping out OpenAI models for its own in-house alternatives — internally branded as MAI — to power productivity staples like Word and Excel. The motivation was straightforward: cost reduction. Running OpenAI inference at Microsoft's scale isn't cheap, and building your own models, even slightly worse ones, starts looking attractive when you're paying per token for hundreds of millions of users.

That reporting kicked off the predictable round of "are Microsoft and OpenAI breaking up?" speculation. Which, honestly, is a reasonable question. The two companies have been sending increasingly mixed signals about the nature of their relationship — a multi-billion-dollar partnership that's starting to look less like a strategic alliance and more like two people who moved in together and are now aggressively splitting the grocery bill.

The Announcement, Translated

So on Thursday, timed to the launch of GPT-5.6, OpenAI published a blog post declaring that the new model would be the "preferred model" powering Microsoft 365 Copilot — including Word, Excel, PowerPoint, and Cowork. The quote they led with: "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."

Lovely. But notice what that sentence doesn't contain: specifics. No SLA guarantees, no exclusivity terms, no word on what percentage of 365 Copilot inference will actually route through GPT-5.6 versus MAI models. "Preferred model" is a marketing phrase, not an architectural commitment.

And here's the thing — nothing in this announcement actually contradicts the Bloomberg reporting. Microsoft using its own MAI models in certain apps while simultaneously designating GPT-5.6 as the "preferred" model for the flagship Copilot product are not mutually exclusive facts. Both can be true. "Preferred" is doing a lot of quiet heavy lifting in that sentence.

Why the Framing Matters

For engineers and builders integrating with the Microsoft 365 ecosystem, this matters in a practical sense. If you're developing Copilot extensions or relying on the quality characteristics of GPT-class models for enterprise workflows, you want to know which inference stack you're actually building on. "Preferred" doesn't tell you whether that's 100% of Copilot calls, 60%, or just the flagship demos.

The MAI question also isn't going away. Microsoft has serious research infrastructure — the folks who built MAI aren't amateurs — and the economic pressure to reduce OpenAI API dependency will only intensify as Microsoft's own model capabilities improve. This isn't betrayal; it's just vertical integration. Every big tech company eventually tries to own its own AI stack.

What the GPT-5.6 announcement does signal, at minimum, is that OpenAI still has the better product for Microsoft's most visible, highest-stakes AI surface area. 365 Copilot is the crown jewel of Microsoft's enterprise AI pitch. Using your best external model there while quietly substituting cheaper in-house models in less scrutinized corners of the product suite is... actually just rational engineering.

The Bigger Picture

The OpenAI-Microsoft relationship has always been structurally weird. Microsoft is simultaneously OpenAI's biggest investor, primary cloud infrastructure provider, and a direct competitor building its own models. That's not a partnership — that's a Jenga tower.

GPT-5.6 landing as the preferred engine for 365 Copilot is genuinely good news for OpenAI's revenue and relevance. But reading this as proof that the two companies are as aligned as ever requires selectively ignoring the cost-cutting story that prompted the announcement in the first place. The relationship is evolving, probably toward something where Microsoft sources strategically from OpenAI while building out its own capabilities in parallel. That's not a breakup. It's just the natural trajectory of any supplier relationship once the buyer gets sophisticated enough to build alternatives.

In the meantime, GPT-5.6 is apparently good enough that Microsoft wants it fronting their most important enterprise product. That's worth something. Just don't confuse the press release for the contract.