OpenAI reinforces its role in Microsoft 365 Copilot as Microsoft continues to push its own MAI models

OpenAI has taken advantage of the expanded release of GPT-5.6 to send a very specific message about its relationship with Microsoft. The company has announced that GPT-5.6 will be the “preferred model” for Microsoft 365 Copilota clarification that comes just after several reports that pointed to greater use by Microsoft of its own MAI models in applications such as Word and Excel with the aim of cutting costs.

In recent days, different publications had indicated that Microsoft would be partially reducing their dependence of OpenAI and Anthropic models in some flows within its office suite, replacing them in certain cases with internal models. In this context, OpenAI wanted to highlight that its technology will continue to occupy a central place in Microsoft 365 Copilot, where GPT-5.6 will support users of Word, Excel, PowerPoint and Cowork.

The announcement also comes shortly after OpenAI launched GPT-5.6 more broadly, its new family of Sol, Terra and Luna models, with a speech very focused on efficiency, safety and operating cost. That nuance is important because part of the recent debate between laboratories and large enterprise clients no longer revolves only around who has the best model, but also how much it costs to keep it in large-scale production.

GPT-5.6 gains visibility as Microsoft tries to gain margin

Obviously, the fact that OpenAI talks about a “preferred model” does not necessarily mean exclusivity. What it does make clear is that Microsoft will continue to rely on GPT-5.6 within Copilot 365, even if in parallel it continues to increase the use of its own systems in certain contexts. Sam Altman’s company has insisted that its alliance with Microsoft continues to be aimed at bringing advanced artificial intelligence to more individuals and organizations, a way of underlining continuity at a time when the market is beginning to read every movement in terms of distancing.

On Microsoft’s part, the development of its own models is not new either. The company has been strengthening its internal line for months and recently presented MAI-Thinking-1 as an example of that progress. However, MAI models are still not up to par with the competition’s frontier models, at least in market perception and positioning compared to the most powerful proposals from OpenAI or Anthropic. It is therefore logical that Microsoft combines its own ambition with selective dependence on external partners.

The underlying question is not so much whether both companies break up, but how they reorder a relationship that can no longer be based on one-way dependency. Microsoft needs more control over cost, integration and roadmapwhile OpenAI needs to maintain its presence in massive enterprise products like Copilot 365. In other words, the two parties still need each other, but each is now trying to reduce risks and gain autonomy at different points in the value chain.

All of this also fits into a broader market trend. Several large technology companies are reviewing the real cost of their bets on artificial intelligence and looking for formulas to contain it, whether through their own models, hybrid combinations or finer adjustments in inference and deployment. In that scenario, the OpenAI announcement works as a sign of strategic continuitybut it does not prevent Microsoft from looking for ways to make the use of AI cheaper in its application ecosystem.