OpenAI says GPT-5.6 is now the “preferred model” in Microsoft 365 Copilot, a move that appears designed to reassure customers and partners that its models remain central to Microsoft’s workplace AI products even as reports suggest Microsoft is leaning more heavily on internal models.

According to OpenAI’s official announcement, GPT-5.6 will support users across Microsoft’s productivity suite, including Word, Excel, PowerPoint, Chat, and Cowork. The timing matters. Earlier this week, Bloomberg, as cited by TechCrunch, reported that Microsoft had been replacing some OpenAI software with in-house models known as MAI to lower costs, especially in apps such as Word and Excel. OpenAI’s new positioning of GPT-5.6 does not directly contradict that reporting, but it does establish that the OpenAI model family still has a formal and visible role inside Microsoft 365 Copilot.

For AI buyers and product teams, the announcement is less about a clean exclusivity arrangement than about how a major enterprise platform is balancing model quality, cost, and strategic control. The phrase “preferred model” is meaningful, but not fully defined in the public materials cited here.

What OpenAI actually announced

The confirmed news from OpenAI is narrow but important: GPT-5.6 is now the preferred model in Microsoft 365 Copilot. In its official post, OpenAI said the model would power experiences across Microsoft 365 apps including Word, Excel, PowerPoint, Chat, and Cowork.

OpenAI also framed the announcement as a continuation of its relationship with Microsoft rather than a new partnership structure. In the company’s published statement, OpenAI said its work with Microsoft has been about bringing advanced AI to more people and organizations and that it plans to keep building on that shared effort.

What the announcement does not clarify is the operational meaning of “preferred model.” Based on the available evidence, OpenAI did not spell out whether GPT-5.6 is the default model for every Microsoft 365 Copilot query, whether it is used only for selected workloads, or whether Microsoft can still route tasks to MAI or other models based on cost, latency, or domain fit.

That ambiguity is central to interpreting the news. Enterprises buying Microsoft 365 Copilot may hear “preferred model” as a quality signal, but the source material does not establish exclusivity, universal routing, or a permanent commitment.

Why the timing matters

The announcement landed against a backdrop of growing scrutiny around the Microsoft and OpenAI relationship. TechCrunch linked the move directly to recent reporting that Microsoft has been increasing its use of MAI, a family of internal models, in products including Word and Excel.

That broader context has fueled recurring speculation that the companies are becoming less tightly linked than they once appeared. For several years, Microsoft Azure, Microsoft 365 Copilot, and OpenAI products such as ChatGPT have been publicly associated in the market’s mind. But as enterprise AI matures, large platform vendors increasingly want multiple options: top-tier external models when quality matters most, and internal models when cost control or tighter product integration becomes more important.

Seen through that lens, OpenAI’s announcement reads as both a product update and a strategic signal. It tells the market that Microsoft 365 Copilot still relies on OpenAI at a high level, even if Microsoft is also building alternatives. That matters because Microsoft 365 Copilot is one of the most visible enterprise AI deployments in the market, and because Microsoft remains a major distribution channel for OpenAI technology in office workflows.

“Preferred” does not mean exclusive

The main unresolved issue is whether GPT-5.6’s new status changes anything material about Microsoft’s model mix.

TechCrunch explicitly noted that prior reporting never said OpenAI models would disappear from Microsoft apps. The claim was that Microsoft had been using its own models more often in order to reduce costs. OpenAI’s new disclosure, as described by TechCrunch, does not appear to negate that. In other words, both things can be true at once: GPT-5.6 can be the preferred model for Microsoft 365 Copilot, while Microsoft still expands the use of MAI where it sees financial or technical advantages.

This distinction matters for enterprise customers trying to evaluate performance consistency. A “preferred model” label may point to the model Microsoft wants associated with flagship Copilot quality. But unless Microsoft or OpenAI provides routing details, customers should not assume that every prompt in Microsoft 365 Copilot is handled by GPT-5.6 under all conditions.

For builders, this is also a reminder that branded AI products are increasingly becoming orchestration layers rather than simple wrappers around a single foundation model. A user may think they are using one assistant, but the underlying system could involve policy-based routing across OpenAI, MAI, or other model families depending on the task.

Evidence, claims, and what remains unverified

The strongest confirmed fact in this story comes from OpenAI News: GPT-5.6 is now the preferred model in Microsoft 365 Copilot, and OpenAI says the model supports experiences across Word, Excel, PowerPoint, Chat, and Cowork.

The market context comes from TechCrunch’s reporting, which cites Bloomberg’s earlier report that Microsoft has been replacing some OpenAI software with in-house MAI models in an effort to cut costs. In the evidence provided here, Bloomberg’s original reporting is secondhand through TechCrunch, so the exact scope of those substitutions is not independently detailed in the source set.

There are also no public technical benchmarks in the supplied evidence showing why GPT-5.6 earned preferred status inside Microsoft 365 Copilot. OpenAI’s official summary says the model offers stronger capabilities for faster, higher-quality work, but without the full official article text or independent testing, those should be treated as vendor claims rather than externally verified performance results.

Likewise, there is no customer usage data in the source material showing how often GPT-5.6 is invoked across Microsoft 365 Copilot or which application features depend on it most. There are also no disclosed commercial terms, no explanation of whether Microsoft Azure remains the exclusive infrastructure layer for these workloads, and no public description of when MAI may be selected instead.

So the evidence supports a specific conclusion, but not a sweeping one: OpenAI remains formally embedded in Microsoft 365 Copilot through GPT-5.6, while the deeper economics and routing architecture remain opaque.

What this means for enterprise buyers and AI builders

For enterprises, the practical takeaway is that Microsoft 365 Copilot is still closely tied to OpenAI quality signaling. If an organization chose Microsoft 365 Copilot partly because it wanted access to OpenAI-grade reasoning and language generation inside office tools, this announcement should be reassuring.

At the same time, cost pressure is clearly part of the story. If Microsoft is indeed using MAI more broadly behind the scenes, that suggests a future in which enterprise AI platforms optimize aggressively for margin and workload type. Buyers should expect more hybrid stacks, not fewer. That can be good for price and latency, but it can also complicate governance, reproducibility, and evaluation if different model families behave differently across tasks.

For product teams building on Microsoft 365 Copilot, GPT-5.6’s preferred status may indicate where Microsoft and OpenAI believe premium user-facing performance matters most: document drafting in Word, analysis in Excel, presentation generation in PowerPoint, and conversational workflows in Chat and Cowork. But teams should avoid assuming that model behavior is static. If Microsoft is balancing OpenAI and MAI, application-level outputs could evolve as routing logic changes.

For the broader enterprise AI market, the episode highlights a structural shift. Big software vendors want the branding power of OpenAI while reducing dependence on any one supplier. Model companies, meanwhile, want to remain the visible intelligence layer inside high-distribution products such as Microsoft 365 Copilot. The result is a partnership model that looks stable on the surface but increasingly modular underneath.

What to watch next

First, watch whether Microsoft defines “preferred model” in more technical or contractual terms. If the company clarifies default routing, fallback logic, or app-by-app model assignment, enterprise buyers will have a better sense of what they are actually purchasing.

Second, watch for evidence of MAI expansion inside Microsoft products beyond what has been reported. If Microsoft continues to push MAI into Word, Excel, or other Copilot experiences, that would support the view that this is becoming a multi-model stack with OpenAI in the flagship role rather than the only role.

Third, look for independent testing of GPT-5.6 inside Microsoft 365 Copilot. Without third-party evaluation, claims about faster or higher-quality work remain primarily vendor-reported.

Finally, pay attention to how this affects Microsoft Azure positioning. If OpenAI’s preferred-model role in Microsoft 365 Copilot is paired with continued infrastructure reliance on Microsoft Azure, that would reinforce the partnership’s commercial depth even as the product layer becomes more diversified.

Creati.ai perspective

The most interesting part of this announcement is not the label itself but what it reveals about enterprise AI distribution. OpenAI wants customers to know that GPT-5.6 still sits at the heart of Microsoft 365 Copilot. Microsoft, if recent reporting is directionally correct, also wants the freedom to use MAI wherever cost or control makes that sensible. Those goals are not mutually exclusive.

For AI builders and buyers, that means the real competition is shifting from single-model supremacy to orchestration advantage. The winner may not be the company with one model everywhere, but the platform that can decide when to use GPT-5.6, when to use MAI, and how to present that complexity as a coherent Copilot experience. OpenAI’s announcement shows it is still inside the most important enterprise productivity surface in AI. It does not, however, settle the deeper question of how much of that surface Microsoft ultimately wants to own itself.