On June 2, 2026, at its Build developer conference in San Francisco, Microsoft AI chief Mustafa Suleyman unveiled seven new in-house MAI models, a clear step toward reducing the company’s dependence on OpenAI. The lineup included MAI-Thinking-1, Microsoft’s first reasoning model, described as having 35 billion active parameters and a 256K context window, which the company says testers preferred to Sonnet 4.6 and which matches Opus 4.6 on coding. It also included MAI-Code-1, an inference-efficient coding model tuned for GitHub and available in Copilot and Visual Studio Code.
The rest of the slate spanned modalities: MAI-Image-2.5 and a flash variant for text-to-image and image-to-image generation (ranked near the top of the Arena leaderboard and shipping in PowerPoint), MAI Transcribe 1.5 for speech recognition across 43 languages, and MAI-Voice-2 plus a flash variant for synthetic voice in 15-plus languages. Microsoft emphasized efficiency and cost across the family.
For business readers, the significance is strategic rather than purely technical. The company that bankrolled OpenAI is now building competitive frontier-class models of its own, signaling that the largest AI buyers increasingly want to own the model layer - a shift that reshapes vendor leverage, pricing, and the structure of the AI market.