Microsoft just made public something it has been building quietly for a while: an AI model it owns completely, top to bottom, with no OpenAI involved anywhere in it.
The model is called MAI-Thinking-1, and it just moved from a closed test group into public preview, meaning any business can now try it through Microsoft's cloud platform, Foundry. It is built for reasoning tasks: solving math problems, writing and fixing computer code, and working through long documents step by step rather than just spitting out a quick answer.
Here is why this matters more than a typical product update. For the last several years, Microsoft's entire AI push, including the version of Copilot in Word, Excel, and Teams, ran on OpenAI's models. Microsoft invested billions into OpenAI and got exclusive access to its technology in return.
That exclusivity ended earlier this year when the two companies redid their deal, giving both sides more room to work independently. MAI-Thinking-1 is Microsoft acting on that new freedom. It is also proof that Microsoft's own AI team can now build something competitive without leaning on its partner.
Microsoft is now, at the same time, OpenAI's biggest investor, OpenAI's biggest customer through Azure cloud services, and OpenAI's direct competitor with its own models. That is an unusual position for any company to be in. It tells you Microsoft does not want to bet its future entirely on one partner, no matter how close that partnership is.
The business case Microsoft is making centers on cost and trust. The model is smaller than many top competitors, which Microsoft says lets it run cheaper per task without giving up much quality on coding and math benchmarks. Microsoft also says the training data behind the model is fully licensed and traceable, which matters if your company worries about copyright risk when using AI output in client work.
A word of caution: the benchmark scores and the claim that human testers preferred this model over Anthropic's Claude Sonnet 4.6 both come from Microsoft's own testing, run with a partner Microsoft chose. No outside lab has independently confirmed these results yet.
What this means for a business buying AI tools today: you now have another serious option beyond OpenAI, Anthropic, and Google, one that plugs directly into Microsoft 365, GitHub Copilot, and Excel if your company already runs on Microsoft's cloud. If you are budgeting for AI tools next year, it is worth testing this model against whatever you use now on your own real work, not on Microsoft's benchmark slides, before switching anything over.
The bigger signal is about where the AI industry is heading. The biggest cloud and software companies no longer want to rent their core intelligence from someone else. Microsoft joins Amazon, Google, and Meta in building its own frontier models rather than depending fully on outside labs, which should mean more competition and more price pressure over time.